optimizer.py 164.4 KB
Newer Older
1
# Copyright (c) 2019 PaddlePaddle Authors. All Rights Reserved.
D
dzhwinter 已提交
2
#
D
dzhwinter 已提交
3 4 5
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
D
dzhwinter 已提交
6
#
D
dzhwinter 已提交
7
#     http://www.apache.org/licenses/LICENSE-2.0
D
dzhwinter 已提交
8
#
D
dzhwinter 已提交
9 10 11 12 13
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
14 15

from __future__ import print_function
16

17
import numpy as np
18
from collections import defaultdict
19

Q
Qiao Longfei 已提交
20
from paddle.fluid.distribute_lookup_table import find_distributed_lookup_table
21
from paddle.fluid.framework import Program, Variable, name_scope, default_main_program, default_startup_program
22

23 24
from . import framework
from . import layers
25
from . import unique_name
26
from .backward import append_backward, _some_in_set_, _append_grad_suffix_, _get_no_grad_set_name
27
from .clip import append_gradient_clip_ops, error_clip_callback
28 29 30
from .framework import program_guard
from .initializer import Constant
from .layer_helper import LayerHelper
S
sneaxiy 已提交
31
from .layers import ops
32
from .regularizer import append_regularization_ops
33
from .dygraph import base as imperative_base
34
from .dygraph import no_grad
35 36 37 38
from .dygraph.learning_rate_scheduler import LearningRateDecay
from paddle.fluid import core
from paddle.fluid.layers import tensor
from functools import reduce
39
from .wrapped_decorator import signature_safe_contextmanager
M
mapingshuo 已提交
40
from .. import compat as cpt
41

42
__all__ = [
43 44 45 46
    'SGD', 'Momentum', 'Adagrad', 'Adam', 'Adamax', 'Dpsgd', 'DecayedAdagrad',
    'Ftrl', 'SGDOptimizer', 'MomentumOptimizer', 'AdagradOptimizer',
    'AdamOptimizer', 'AdamaxOptimizer', 'DpsgdOptimizer',
    'DecayedAdagradOptimizer', 'RMSPropOptimizer', 'FtrlOptimizer', 'Adadelta',
Z
Zeng Jinle 已提交
47 48 49 50
    'AdadeltaOptimizer', 'ModelAverage', 'LarsMomentum',
    'LarsMomentumOptimizer', 'DGCMomentumOptimizer', 'LambOptimizer',
    'ExponentialMovingAverage', 'PipelineOptimizer', 'LookaheadOptimizer',
    'RecomputeOptimizer'
51
]
Q
Qiao Longfei 已提交
52 53 54 55 56 57


class Optimizer(object):
    """Optimizer Base class.

    Define the common interface of an optimizer.
58 59
    User should not use this class directly,
    but need to use one of it's implementation.
Q
Qiao Longfei 已提交
60 61
    """

62
    @imperative_base.no_grad
63 64 65 66 67 68
    def __init__(self,
                 learning_rate,
                 parameter_list=None,
                 regularization=None,
                 name=None):
        self._parameter_list = None
L
lujun 已提交
69
        if framework.in_dygraph_mode():
M
minqiyang 已提交
70 71 72 73 74
            if not isinstance(learning_rate, float) and \
                    not isinstance(learning_rate, LearningRateDecay):
                raise TypeError(
                    "learning rate should be float or LearningRateDecay, got %s here"
                    % type(learning_rate))
75 76 77 78
            if name is not None:
                self._name = unique_name.generate(name)
            else:
                self._name = unique_name.generate(self.__class__.__name__)
79 80 81 82 83 84
            if parameter_list is not None:
                self._parameter_list = parameter_list
            else:
                raise AttributeError(
                    "parameter_list argument given to the Optimizer should not be None in dygraph mode."
                )
M
minqiyang 已提交
85 86 87 88 89 90
        else:
            if not isinstance(learning_rate, float) and \
                    not isinstance(learning_rate, framework.Variable):
                raise TypeError(
                    "learning rate should be float or Variable, got %s here" %
                    type(learning_rate))
91
            self._name = name
M
minqiyang 已提交
92

D
dzhwinter 已提交
93
        self.regularization = regularization
94
        self._learning_rate = learning_rate
D
dzhwinter 已提交
95 96
        # the learning rate type should be inferenced from loss
        self._dtype = None
97
        # each program should have a independent learning rate
98
        # program -> Variable(learning_rate)
Q
qiaolongfei 已提交
99
        self._learning_rate_map = dict()
100
        if isinstance(self._learning_rate, framework.Variable):
101 102
            self._learning_rate_map[framework.default_main_program(
            )] = self._learning_rate
103 104 105 106 107
        # Dictionary of accumulators. Some optimizer subclasses need to
        # allocate and manage extra variables associated with the parameters
        # to train. These variables are called accumulators.
        # {accum_name : { paramter_name : accumulator_for_parameter, ...}, ...}
        self._accumulators = defaultdict(lambda: dict())
Q
Qiao Longfei 已提交
108
        self.helper = None
109
        self._opti_name_list = []
H
hong 已提交
110 111 112 113 114
        self._accumulators_holder = {}

    @framework.dygraph_only
    def state_dict(self):
        '''
T
tianshuo78520a 已提交
115 116
        Get state dict information from optimizer. It contain all the variable used by optimizer. For Adam optimizer, contains beta1, beta2, momentum etc. If LearningRateDecay have been used, global_step will be include in state dict.
        If the optimizer never be called(minimize function), the state_dict is empty.
H
hong 已提交
117 118 119

        Args: None
        Return:
T
tianshuo78520a 已提交
120
            state_dict(dict) : dict contains all the variable used by optimizer
H
hong 已提交
121 122 123 124 125
        
        Examples:
            .. code-block:: python

                import paddle.fluid as fluid
126 127 128 129 130 131

                with fluid.dygraph.guard():
                    emb = fluid.dygraph.Embedding([10, 10])

                    adam = fluid.optimizer.Adam(0.001, parameter_list=emb.parameters())
                    state_dict = adam.state_dict()
H
hong 已提交
132 133 134 135 136 137 138 139

        '''
        state_dict = {}
        for k, v in self._accumulators.items():
            for para_name, var_tmp in v.items():
                state_dict[var_tmp.name] = var_tmp
        # global step if use lr decay
        if isinstance(self._learning_rate, LearningRateDecay):
140 141 142 143 144 145 146
            var_tmp = None
            if not framework.in_dygraph_mode():
                var_temp = Variable(None, name='global_step', dtype='int32')
            else:
                var_temp = framework._varbase_creator(
                    None, name='global_step', dtype='int32')

H
hong 已提交
147 148 149 150 151 152 153 154 155
            tensor.fill_constant(
                [1], "int32", self._learning_rate.step_num, out=var_temp)

            state_dict['global_step'] = var_temp
        return state_dict

    @framework.dygraph_only
    def set_dict(self, state_dict):
        '''
T
tianshuo78520a 已提交
156
        Load optimizer state dict. For Adam optimizer, contains beta1, beta2, momentum etc. If LearningRateDecay have been used, global_step will be changed.
H
hong 已提交
157 158 159 160 161 162 163 164

        Args: 
            state_dict(dict) : Dict contains all the Variable needed by optimizer
        Return:
            None
        
        Examples:
            .. code-block:: python
165

H
hong 已提交
166
                with fluid.dygraph.guard():
167
                    emb = fluid.dygraph.Embedding([10, 10])
168

H
hong 已提交
169
                    state_dict = emb.state_dict()
170
                    fluid.save_dygraph(state_dict, "paddle_dy")
171

172 173
                    adam = fluid.optimizer.Adam(learning_rate=fluid.layers.noam_decay( 100, 10000), 
                                                parameter_list=emb.parameters())
H
hong 已提交
174
                    state_dict = adam.state_dict()
175
                    fluid.save_dygraph(state_dict, "paddle_dy")
176

H
hong 已提交
177
                    para_state_dict, opti_state_dict = fluid.load_dygraph( "paddle_dy")
178

179
                    adam.set_dict(opti_state_dict)
H
hong 已提交
180 181 182 183 184 185 186 187 188

        '''

        if isinstance(self._learning_rate, LearningRateDecay):
            assert 'global_step' in state_dict, \
                    'Global step not in state dict, Dygraph use LearningRateDecay, global_step must in state_dict'
            global_step = state_dict['global_step']

            if isinstance(global_step, core.VarBase):
189
                step_np = global_step
H
hong 已提交
190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213
                step_np = np.array(step_np.value().get_tensor())
                assert step_np.shape == (1,),  \
                        "global step shape is (1,), the shape is {}".format( step_np.shape )

                self._learning_rate.step_num = int(step_np[0])
            elif isinstance(global_step, Variable):
                step_np = global_step.numpy()
                assert step_np.shape == (1,),  \
                        "global step shape is (1,), the shape is {}".format( step_np.shape )
                self._learning_rate.step_num = step_np[0]
            elif isinstance(global_step, np.ndarray):
                assert global_step.shape == (1,),  \
                        "global step shape is (1,), the shape is {}".format( global_step.shape )
                self._learning_rate.step_num = global_step[0]
            else:
                raise RuntimeError(
                    "Type not supprt, value in state dict must be [VarBase, Variable, numpy], the type is ",
                    type(global_step))

        self._accumulators_holder = state_dict
        for k, v in self._accumulators.items():
            for para_name, var_tmp in v.items():
                assert var_tmp.name in state_dict, \
                        "optimizer variable {} not found".format( var_tmp.name )
214
                var = var_tmp.value()
H
hong 已提交
215 216 217 218 219 220 221 222
                tensor = var.get_tensor()
                model_np = np.array(tensor)

                load_para = state_dict[var_tmp.name]

                if isinstance(load_para, Variable):
                    load_para_np = load_para.numpy()
                elif isinstance(load_para, core.VarBase):
223
                    load_para_np = load_para.numpy()
H
hong 已提交
224 225 226 227 228 229 230 231 232 233 234 235 236 237 238
                elif isinstance(load_para, np.ndarray):
                    load_para_np = load_para
                else:
                    raise RuntimeError("State dict type {} not supprt".format(
                        str(type(load_para))))

                assert model_np.shape == load_para_np.shape,  \
                                          "Parameter shape not match, Dygraph Parameter [ {} ] need tensor with shape {} but load tensor with shape {}".format(
                                                 item.name, model_np.shape, load_para_np.shape)

                assert model_np.dtype == load_para_np.dtype, \
                                          "Parameter dtype not match, Dygraph Parameter [ {} ] need tensor with dtype {}  but load tensor with dtype {}".format(
                                                item.name, model_np.dtype, load_para_np.dtype)

                tensor.set(load_para_np, framework._current_expected_place())
239

240 241
    def get_opti_var_name_list(self):
        return self._opti_name_list
Q
Qiao Longfei 已提交
242

Q
Qiao Longfei 已提交
243
    def _create_global_learning_rate(self):
244 245 246
        if imperative_base.enabled():
            # create learning rate Variable
            if isinstance(self._learning_rate, float):
M
minqiyang 已提交
247 248 249 250 251 252 253 254 255 256 257 258
                lr = self._global_learning_rate()

                if isinstance(lr, framework.Variable):
                    return
                else:
                    self._learning_rate_map[framework.default_main_program(
                    )] = layers.create_global_var(
                        name=unique_name.generate("learning_rate"),
                        shape=[1],
                        value=float(self._learning_rate),
                        dtype='float32' if self._dtype is None else self._dtype,
                        persistable=True)
259
            # get learning rate Variable from LearningRateDecay
M
minqiyang 已提交
260
            elif isinstance(self._learning_rate, LearningRateDecay):
261 262 263
                self._learning_rate_map[framework.default_main_program(
                )] = self._learning_rate()
            else:
Q
qiaolongfei 已提交
264
                raise TypeError(
265 266
                    "optimizer's learning rate must be float or LearningRateDecay"
                )
267
        else:
268 269 270 271
            lr = self._global_learning_rate()

            if isinstance(lr, framework.Variable):
                return
M
minqiyang 已提交
272 273 274 275 276 277
            else:
                if not isinstance(self._learning_rate, float):
                    raise TypeError(
                        "learning rate variable is create outside optimizer,"
                        "can not create new learning rate variable for new program"
                    )
Q
Qiao Longfei 已提交
278

279 280 281 282 283 284 285 286
            # create learning rate in the current main program
            self._learning_rate_map[framework.default_main_program(
            )] = layers.create_global_var(
                name=unique_name.generate("learning_rate"),
                shape=[1],
                value=float(self._learning_rate),
                dtype='float32' if self._dtype is None else self._dtype,
                persistable=True)
287

288 289 290 291
    @framework.dygraph_only
    def current_step_lr(self):
        """
        .. note::
T
tianshuo78520a 已提交
292
          **This API is ONLY available in Dygraph mode**
293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349
        
        Get current step learning rate. The return value is all the same When LearningRateDecay is not used,
        otherwise return the step learning rate.

        Returns:
            float: The learning rate of the current step.

        Examples:
            .. code-block:: python

                import paddle.fluid as fluid
                import numpy as np

                # example1: LearningRateDecay is not used, return value is all the same
                with fluid.dygraph.guard():
                    emb = fluid.dygraph.Embedding([10, 10])
                    adam = fluid.optimizer.Adam(0.001, parameter_list = emb.parameters())
                    lr = adam.current_step_lr()
                    print(lr) # 0.001

                # example2: PiecewiseDecay is used, return the step learning rate
                with fluid.dygraph.guard():
                    inp = np.random.uniform(-0.1, 0.1, [10, 10]).astype("float32")
                    linear = fluid.dygraph.nn.Linear(10, 10)
                    inp = fluid.dygraph.to_variable(inp)
                    out = linear(inp)
                    loss = fluid.layers.reduce_mean(out)
                    
                    bd = [2, 4, 6, 8]
                    value = [0.2, 0.4, 0.6, 0.8, 1.0]
                    adam = fluid.optimizer.Adam(fluid.dygraph.PiecewiseDecay(bd, value, 0),
                                           parameter_list=linear.parameters())

                    # first step: learning rate is 0.2
                    np.allclose(adam.current_step_lr(), 0.2, rtol=1e-06, atol=0.0) # True

                    # learning rate for different steps
                    ret = [0.2, 0.2, 0.4, 0.4, 0.6, 0.6, 0.8, 0.8, 1.0, 1.0, 1.0, 1.0]
                    for i in range(12):
                        adam.minimize(loss)
                        lr = adam.current_step_lr()
                        np.allclose(lr, ret[i], rtol=1e-06, atol=0.0) # True

        """
        current_lr = self._global_learning_rate()
        if current_lr:
            return self._global_learning_rate().numpy()[0]

        if isinstance(self._learning_rate, float):
            return self._learning_rate
        else:
            step_lr = self._learning_rate.step()
            if isinstance(step_lr, (float, int)):
                return step_lr
            else:
                return step_lr.numpy()[0]

Y
yuyang18 已提交
350
    def _global_learning_rate(self, program=None):
Q
Qiao Longfei 已提交
351 352 353 354
        """
        get global decayed learning rate
        :return:
        """
355 356
        if program is None:
            program = framework.default_main_program()
Q
qiaolongfei 已提交
357
        return self._learning_rate_map.get(program, None)
Q
Qiao Longfei 已提交
358

Q
Qiao Longfei 已提交
359 360 361 362 363
    def _append_optimize_op(self, block, param_and_grad):
        """ append optimize operator to block and return all the added optimize_op
        """
        raise NotImplementedError()

364 365 366 367
    def _create_param_lr(self, param_and_grad):
        # create learning rate variable for every parameter
        param = param_and_grad[0]
        param_lr = param.optimize_attr['learning_rate']
W
Wu Yi 已提交
368 369
        if type(param_lr) == Variable:
            return param_lr
Q
qiaolongfei 已提交
370
        else:
W
Wu Yi 已提交
371
            if param_lr == 1.0:
Y
yuyang18 已提交
372
                return self._global_learning_rate()
W
Wu Yi 已提交
373
            else:
X
Xin Pan 已提交
374 375 376
                with default_main_program()._lr_schedule_guard(
                        is_with_opt=True), framework.name_scope(
                            'scale_with_param_lr'):
377
                    return self._global_learning_rate() * param_lr
378 379 380 381 382 383 384

    def _create_accumulators(self, block, parameters):
        """Create all accumulators needed by the parameters

        Args:
            block: the block in which the loss variable is present
            parameters: list of parameter variables for the optimizer
Q
Qiao Longfei 已提交
385
        """
386 387
        pass

388
    def _finish_update(self, block, parameters_and_grads):
389 390 391 392 393 394 395 396
        """Finish any custom updates needed
           before completing an optimization step

        Args:
            block: the block in which the loss variable is present
            parameters: list of parameter variables for the optimizer

        Returns:
Q
qiaolongfei 已提交
397
            None
398 399 400
        """
        pass

401 402 403 404 405
    def _add_accumulator(self,
                         name,
                         param,
                         dtype=None,
                         fill_value=0.0,
406 407
                         shape=None,
                         type=None):
408 409 410 411 412 413 414 415 416
        """Utility function to add an accumulator for a parameter

        Args:
            block: the block in which the loss variable is present
            name: name of the accumulator
            param: parameter variable for which accumulator is to be added
            dtype: data type of the accumulator variable
            fill_value: value to initialize the accumulator variable
        """
W
whs 已提交
417 418
        if self._name is not None:
            name = self._name + "_" + name
419 420
        if (name in self._accumulators and
                param.name in self._accumulators[name]):
L
lujun 已提交
421
            if framework.in_dygraph_mode():
X
polish  
Xin Pan 已提交
422
                return self._accumulators[name][param.name]
423
            raise Exception("Accumulator {} already exists for parameter {}".
424
                            format(name, param.name))
425 426
        if shape == None:
            shape = param.shape
Q
Qiao Longfei 已提交
427
        assert isinstance(self.helper, LayerHelper)
428 429 430 431 432

        var_name = param.name + "_" + name
        var_name = unique_name.generate(var_name)
        self._opti_name_list.append(var_name)

Q
Qiao Longfei 已提交
433
        var = self.helper.create_global_variable(
434
            name=var_name,
Q
Qiao Longfei 已提交
435
            persistable=True,
F
fengjiayi 已提交
436
            dtype=dtype or param.dtype,
437
            type=param.type if type is None else type,
H
hong 已提交
438 439
            shape=shape,
            belong_to_optimizer=True)
Q
Qiao Longfei 已提交
440
        self.helper.set_variable_initializer(
441
            var, initializer=Constant(value=float(fill_value)))
H
hong 已提交
442 443 444 445 446 447 448

        if framework.in_dygraph_mode():
            if len(self._accumulators_holder) > 0:
                assert var_name in self._accumulators_holder, \
                        "Optimizer set error, {} should in state dict".format( var_name )
                var.set_value(self._accumulators_holder[var_name])

Q
Qiao Longfei 已提交
449
        self._accumulators[name][param.name] = var
450
        return var
451 452 453 454 455 456 457 458 459 460 461

    def _get_accumulator(self, name, param):
        """Utility function to fetch an accumulator for a parameter

        Args:
            name: name of the accumulator
            param: parameter variable for which accumulator is to be fetched

        Returns:
            accumulator variable for the parameter
        """
W
whs 已提交
462 463
        if self._name is not None:
            name = self._name + "_" + name
464 465 466 467 468 469
        if (name not in self._accumulators or
                param.name not in self._accumulators[name]):
            raise Exception("Accumulator {} does not exist for parameter {}".
                            format(name, param.name))
        return self._accumulators[name][param.name]

470
    def _create_optimization_pass(self, parameters_and_grads):
Q
Qiao Longfei 已提交
471 472 473
        """Add optimization operators to update gradients to variables.

        Args:
Q
qiaolongfei 已提交
474
          parameters_and_grads(list(tuple(Variable, Variable))):
475
            a list of (variable, gradient) pair to update.
Q
Qiao Longfei 已提交
476 477

        Returns:
478
          return_op_list: a list of operators that will complete one step of
479 480 481
            optimization. This will include parameter update ops, global step
            update ops and any other custom ops required by subclasses to manage
            their internal state.
Q
Qiao Longfei 已提交
482
        """
483 484 485 486 487
        # This is a default implementation of create_optimization_pass that
        # can be shared by most optimizers. This implementation assumes that
        # the subclass will implement the _append_optimize_op method and the
        #  _initialize_tensors method. The subclass can extend the
        # _create_accumulators method if it needs to create accumulators
488
        # for parameters and extend _finish_update method to add custom ops.
489

490
        # Allways called under program_guard use global block as loss block
491 492 493
        # But if current block is in control flow, append optimize op in the
        # grad block of current block

494
        global_block = framework.default_main_program().global_block()
495 496 497 498 499 500 501 502 503
        target_block = global_block
        current_block = framework.default_main_program().current_block()
        if current_block.idx != global_block.idx:
            assert current_block.backward_block_idx != -1, \
                "current block is not global_block, but it doesn't have backward block."
            target_block = framework.default_main_program().blocks[
                current_block.backward_block_idx]

        start = len(target_block.ops)
504
        self.helper = LayerHelper(self.__class__.__name__)
C
chengduo 已提交
505
        self._create_accumulators(
506
            target_block,
C
chengduo 已提交
507
            [p[0] for p in parameters_and_grads if p[0].trainable])
508 509 510
        self._create_global_learning_rate()

        optimize_ops = []
M
minqiyang 已提交
511
        if framework.in_dygraph_mode():
512 513 514 515 516 517
            for param_and_grad in parameters_and_grads:
                if param_and_grad[1] is None:
                    continue
                with param_and_grad[0].block.program._optimized_guard(
                        param_and_grad):
                    if param_and_grad[0].trainable is True:
518
                        optimize_op = self._append_optimize_op(target_block,
519 520 521 522 523 524 525 526 527
                                                               param_and_grad)
                        optimize_ops.append(optimize_op)
        else:
            for param_and_grad in parameters_and_grads:
                if param_and_grad[1] is None:
                    continue
                with param_and_grad[0].block.program._optimized_guard(
                        param_and_grad), name_scope("optimizer"):
                    if param_and_grad[0].trainable is True:
528
                        optimize_op = self._append_optimize_op(target_block,
529 530
                                                               param_and_grad)
                        optimize_ops.append(optimize_op)
531 532 533

        # Get custom finish ops for subclasses
        # FIXME: Need to fix this once we figure out how to handle dependencies
534
        self._finish_update(target_block, parameters_and_grads)
535

536 537
        end = len(target_block.ops)
        return target_block._slice_ops(start, end)
538 539

    def _process_distribute_lookuptable(self, param_grads):
Q
Qiao Longfei 已提交
540 541 542 543 544 545 546 547 548
        """
        Because distribute lookup table only support SGD optimizer for now, not support
        other optimizer and regularization, so we should find the table parameter out,
        and avoid to add regularization and other op for it, and add sgd optimize op
        for it independently.
        :param param_grads(list((Var, Var))): list of (param, grad) pair.
        :param loss: the loss variable.
        :param startup_program: the startup program
        """
549 550
        program = framework.default_main_program()
        global_block = framework.default_main_program().global_block()
Q
Qiao Longfei 已提交
551 552 553 554 555 556 557 558 559 560 561 562 563 564 565
        table_name = find_distributed_lookup_table(program)
        table_param = None
        table_grad = None
        new_param_grads = []
        for p, g in param_grads:
            if p.name == table_name:
                if table_param is not None:
                    raise RuntimeError(
                        "multi dist table var found, only support one now!")
                table_param = p
                table_grad = g
            else:
                new_param_grads.append((p, g))
        sgd_op = None
        if table_param is not None:
566 567 568 569 570 571 572 573 574 575 576 577 578
            param_and_grad = [table_param, table_grad]
            with table_param.block.program._optimized_guard(param_and_grad), \
                    framework.name_scope("optimizer"):
                self._create_global_learning_rate()
                # create the optimize op
                sgd_op = global_block.append_op(
                    type='sgd',
                    inputs={
                        "Param": table_param,
                        "Grad": table_grad,
                        "LearningRate": self._create_param_lr(param_and_grad)
                    },
                    outputs={"ParamOut": param_and_grad[0]})
Q
Qiao Longfei 已提交
579 580
        return new_param_grads, (table_param, table_grad), sgd_op

581 582 583
    def _append_dgc_ops(self, param_and_grad):
        pass

584 585 586 587 588 589 590
    def backward(self,
                 loss,
                 startup_program=None,
                 parameter_list=None,
                 no_grad_set=None,
                 callbacks=None):
        """
591
        The first part of ``minimize``, do auto-diff to append backward operations for
592 593 594
        the current program.

        Args:
595 596 597 598
            loss (Variable): ``loss`` variable to run optimizations.
            startup_program (Program, optional): :ref:`api_fluid_Program` for
                initializing parameters in ``parameter_list``. The default value
                is None, at this time :ref:`api_fluid_default_startup_program` will be used.
599
            parameter_list (list, optional): List of ``Variable`` or ``Variable.name`` to update
600 601
                to minimize ``loss``. The default value is None, at this time all parameters
                will be updated.
602
            no_grad_set (set, optional): Set of ``Variable``  or ``Variable.name`` that don't need
603 604 605
                to be updated. The default value is None.
            callbacks (list, optional): list of callable objects to run when appending backward
                operator for one parameter. The default value is None.
M
minqiyang 已提交
606

607
        Return:
608 609
            list: list of (param, grad) variable pairs, param is ``Parameter``,
                grad is the gradient value corresponding to the parameter.
M
minqiyang 已提交
610

611
        Examples:
612
            See examples in ``apply_gradients``.
613
        """
614 615 616 617 618
        act_no_grad_set = None
        if not framework.in_dygraph_mode():
            act_no_grad_set = self._get_no_grad_set(loss, no_grad_set)
        else:
            pass
G
gongweibao 已提交
619

C
chengduo 已提交
620
        self._dtype = loss.dtype
L
lujun 已提交
621
        if framework.in_dygraph_mode():
C
chengduo 已提交
622
            params_grads = []
623
            for param in self._parameter_list:
C
chengduo 已提交
624 625
                if not param.trainable:
                    continue
626
                if param._grad_ivar() is not None:
C
chengduo 已提交
627
                    # create gradient variable
628
                    grad_var = param._grad_ivar()
C
chengduo 已提交
629
                    params_grads.append((param, grad_var))
630
        else:
C
chengduo 已提交
631 632 633 634 635
            if callbacks is None:
                callbacks = [error_clip_callback]
            else:
                assert (isinstance(callbacks, list))
            program = loss.block.program
C
chengduo 已提交
636 637 638 639
            assert len(loss.shape) == 1 and loss.shape[0] == 1, \
                "The loss.shape should be (1L,), but the current loss.shape is {}. " \
                "Maybe that you should call fluid.layers.mean to process the current loss.".format(
                    loss.shape)
C
chengduo 已提交
640 641
            with program_guard(program, startup_program):
                params_grads = append_backward(loss, parameter_list,
642
                                               act_no_grad_set, callbacks)
C
chengduo 已提交
643 644 645 646
                # Note: since we can't use all_reduce_op now,
                #  dgc_op should be the last op of one grad.
                self._append_dgc_ops(params_grads)
        return params_grads
647 648 649 650 651 652 653 654

    def apply_gradients(self, params_grads):
        """
        Second part of `minimize`, appending optimization operators for
        given `params_grads` pairs.

        Args:
            params_grads (list): list of (param, grad) pair to do optimization.
M
minqiyang 已提交
655

656 657
        Returns:
            list: A list of operators appended to the current program.
M
minqiyang 已提交
658

659 660 661
        Examples:
            .. code-block:: python

662
                import paddle.fluid as fluid
663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687
                loss = network()
                optimizer = fluid.optimizer.SGD(learning_rate=0.1)
                params_grads = optimizer.backward(loss)
                # you may append operations for params_grads here
                # ...
                optimizer.apply_gradients(params_grads)
        """
        params_grads = sorted(params_grads, key=lambda x: x[0].name)

        params_grads, table_param_and_grad, table_optimize_op = \
            self._process_distribute_lookuptable(params_grads)

        params_grads = append_gradient_clip_ops(params_grads)

        # Add regularization if any
        params_grads = append_regularization_ops(params_grads,
                                                 self.regularization)

        optimize_ops = self._create_optimization_pass(params_grads)
        if table_optimize_op is not None:
            optimize_ops.append(table_optimize_op)
            params_grads.append(table_param_and_grad)

        return optimize_ops

C
chengduo 已提交
688 689 690 691 692 693 694 695 696 697 698 699 700 701
    def apply_optimize(self, loss, startup_program, params_grads):
        """
        Second part of `minimize`, appending optimization operators for
        given `params_grads` pairs.

        Args:
            loss (Variable): loss variable to run optimizations.
            startup_program (Program): startup_program for initializing parameters
                in `parameter_list`.
            params_grads (list): list of (param, grad) pair to do optimization.

        Returns:
            list: A list of operators appended to the current program.
        """
L
lujun 已提交
702
        if framework.in_dygraph_mode():
C
chengduo 已提交
703 704
            with program_guard(framework.default_main_program(),
                               framework.default_startup_program()):
705 706
                params_grads = append_regularization_ops(params_grads,
                                                         self.regularization)
C
chengduo 已提交
707 708 709 710 711 712 713
                optimize_ops = self._create_optimization_pass(params_grads)
        else:
            program = loss.block.program
            with program_guard(program, startup_program):
                optimize_ops = self.apply_gradients(params_grads)
        return optimize_ops

G
gongweibao 已提交
714
    def _get_no_grad_set(self, loss, no_grad_set=None):
715
        no_grad_set = _get_no_grad_set_name(no_grad_set)
G
gongweibao 已提交
716 717 718 719 720 721 722 723
        parameters = loss.block.program.global_block().all_parameters()
        param_no_trainable = set(
            [param.name for param in parameters if param.trainable is False])
        # If the parameter is no trainable, it should not have a gradient.
        no_grad_set.update(param_no_trainable)

        return no_grad_set

724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754
    @framework.dygraph_only
    def clear_gradients(self):
        """
        Clear the gradients of all optimized parameters for model.
        
        Returns:
            None
        
        Examples:
            .. code-block:: python

                import paddle.fluid as fluid
                import numpy as np

                with fluid.dygraph.guard():
                    value = np.arange(26).reshape(2, 13).astype("float32")
                    a = fluid.dygraph.to_variable(value)
                    linear = fluid.Linear(13, 5, dtype="float32")
                    # This can be any optimizer supported by dygraph.
                    adam = fluid.optimizer.Adam(learning_rate = 0.01, 
                                                parameter_list = linear.parameters())
                    out = linear(a)
                    out.backward()
                    adam.minimize(out)
                    adam.clear_gradients()

        """
        for p in self._parameter_list:
            if p.trainable:
                p.clear_gradient()

755
    @imperative_base.no_grad
Q
Qiao Longfei 已提交
756 757
    def minimize(self,
                 loss,
758
                 startup_program=None,
Q
Qiao Longfei 已提交
759
                 parameter_list=None,
760 761
                 no_grad_set=None,
                 grad_clip=None):
762
        """
763
        Add operations to minimize ``loss`` by updating ``parameter_list``.
M
minqiyang 已提交
764

765
        Args:
766 767 768 769
            loss (Variable): A ``Variable`` containing the value to minimize.
            startup_program (Program, optional): :ref:`api_fluid_Program` for
                initializing parameters in ``parameter_list``. The default value
                is None, at this time :ref:`api_fluid_default_startup_program` will be used.
770
            parameter_list (list, optional): List of ``Variable`` or ``Variable.name`` to update
771 772
                to minimize ``loss``. The default value is None, at this time all parameters
                will be updated.
773
            no_grad_set (set, optional): Set of ``Variable``  or ``Variable.name`` that don't need
774 775 776 777 778
                to be updated. The default value is None.
            grad_clip (GradClipBase, optional) : Gradient clipping strategy, static
                graph mode does not need to use this argument. Currently, this argument
                only supports gradient clipping in dygraph mode. In the future, this
                argument my be adjusted. The default value is None.
Q
Qiao Longfei 已提交
779

780
        Returns:
781 782 783 784 785 786
            tuple: tuple (optimize_ops, params_grads), A list of operators appended
            by minimize and a list of (param, grad) variable pairs, param is
            ``Parameter``, grad is the gradient value corresponding to the parameter.

        Examples:
            Please refer to the example of current Optimizer.
Q
Qiao Longfei 已提交
787
        """
C
chengduo 已提交
788
        assert isinstance(loss, Variable), "The loss should be an Variable."
C
chengduo 已提交
789 790 791 792 793
        params_grads = self.backward(
            loss,
            startup_program=startup_program,
            parameter_list=parameter_list,
            no_grad_set=no_grad_set)
794 795 796 797 798

        if grad_clip is not None and framework.in_dygraph_mode():
            # TODO(hongyu): FIX later, this is only for dygraph, should be work for static mode
            params_grads = grad_clip(params_grads)

C
chengduo 已提交
799 800
        optimize_ops = self.apply_optimize(
            loss, startup_program=startup_program, params_grads=params_grads)
M
minqiyang 已提交
801

Q
Qiao Longfei 已提交
802
        return optimize_ops, params_grads
Q
Qiao Longfei 已提交
803 804 805


class SGDOptimizer(Optimizer):
Q
qiaolongfei 已提交
806 807 808 809 810 811 812
    """
    Optimizer of the stochastic gradient descent algorithm.

    .. math::

        param\_out = param - learning\_rate * grad

813 814 815
    Parameters:
        learning_rate (float|Variable): The learning rate used to update parameters. \
            Can be a float value or a Variable with one float value as data element.
816 817 818
        parameter_list (list, optional):  List of ``Variable`` names to update to minimize ``loss``. \
            This parameter is required in dygraph mode. \
            The default value is None in static mode, at this time all parameters will be updated.
819 820 821 822
        regularization: A Regularizer, such as :ref:`api_fluid_regularizer_L2DecayRegularizer`. \
            Optional, default is None.
        name (str, optional): This parameter is used by developers to print debugging information. \
            For details, please refer to :ref:`api_guide_Name`. Default is None.
Q
qiaolongfei 已提交
823 824 825 826

    Examples:
        .. code-block:: python

827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851
            import paddle
            import paddle.fluid as fluid
            import numpy as np

            place = fluid.CPUPlace()
            main = fluid.Program()
            with fluid.program_guard(main):
                x = fluid.layers.data(name='x', shape=[13], dtype='float32')
                y = fluid.layers.data(name='y', shape=[1], dtype='float32')
                y_predict = fluid.layers.fc(input=x, size=1, act=None)
                cost = fluid.layers.square_error_cost(input=y_predict, label=y)
                avg_cost = fluid.layers.mean(cost)

                sgd_optimizer = fluid.optimizer.SGD(learning_rate=0.001)
                sgd_optimizer.minimize(avg_cost)

                fetch_list = [avg_cost]
                train_reader = paddle.batch(
                    paddle.dataset.uci_housing.train(), batch_size=1)
                feeder = fluid.DataFeeder(place=place, feed_list=[x, y])
                exe = fluid.Executor(place)
                exe.run(fluid.default_startup_program())
                for data in train_reader():
                    exe.run(main, feed=feeder.feed(data), fetch_list=fetch_list)

Q
Qiao Longfei 已提交
852 853
    """

854 855 856 857 858
    def __init__(self,
                 learning_rate,
                 parameter_list=None,
                 regularization=None,
                 name=None):
Q
Qiao Longfei 已提交
859
        assert learning_rate is not None
Q
Qiao Longfei 已提交
860
        super(SGDOptimizer, self).__init__(
X
Xin Pan 已提交
861
            learning_rate=learning_rate,
862
            parameter_list=parameter_list,
X
Xin Pan 已提交
863 864
            regularization=regularization,
            name=name)
Q
Qiao Longfei 已提交
865 866
        self.type = "sgd"

867
    @no_grad
868
    def _append_optimize_op(self, block, param_and_grad):
869 870 871 872 873 874 875 876 877 878
        if framework.in_dygraph_mode():
            inputs = {
                "Param": [param_and_grad[0]],
                "Grad": [param_and_grad[1]],
                "LearningRate": [self._create_param_lr(param_and_grad)]
            }
            attrs = {}
            outputs = {'ParamOut': [param_and_grad[0]]}
            outs = core.ops.sgd(inputs, attrs, outputs)
            return outs['ParamOut'][0]
879

880
        assert isinstance(block, framework.Block)
Q
Qiao Longfei 已提交
881 882 883 884 885 886
        # create the optimize op
        sgd_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
887
                "LearningRate": self._create_param_lr(param_and_grad)
Q
Qiao Longfei 已提交
888
            },
M
minqiyang 已提交
889 890
            outputs={"ParamOut": param_and_grad[0]},
            stop_gradient=True)
Q
Qiao Longfei 已提交
891 892

        return sgd_op
893 894 895


class MomentumOptimizer(Optimizer):
Q
qiaolongfei 已提交
896 897 898 899 900 901 902 903 904 905 906 907 908 909
    """

    Simple Momentum optimizer with velocity state

    This optimizer has a flag for Nestrov Momentum.

    The update equations are as follows:

    .. math::

        & velocity = mu * velocity + gradient

        & if (use\_nesterov):

910
        &\quad   param = param - (gradient + mu * velocity) * learning\_rate
Q
qiaolongfei 已提交
911 912 913

        & else:

Q
qiaolongfei 已提交
914
        &\quad   param = param - learning\_rate * velocity
Q
qiaolongfei 已提交
915

916 917 918 919
    Parameters:
        learning_rate (float|Variable): The learning rate used to update parameters. \
            Can be a float value or a Variable with one float value as data element.
        momentum (float): Momentum factor
920 921 922
        parameter_list (list, optional):  List of ``Variable`` names to update to minimize ``loss``. \
            This parameter is required in dygraph mode. \
            The default value is None in static mode, at this time all parameters will be updated.
923 924 925 926 927
        use_nesterov (bool, optional): Enables Nesterov momentum, default is false.
        regularization: A Regularizer, such as :ref:`api_fluid_regularizer_L2DecayRegularizer`. \
            Optional, default is None.
        name (str, optional): This parameter is used by developers to print debugging information. \
            For details, please refer to :ref:`api_guide_Name`. Default is None.
Q
qiaolongfei 已提交
928 929 930 931

    Examples:
        .. code-block:: python

932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956
            import paddle
            import paddle.fluid as fluid
            import numpy as np

            place = fluid.CPUPlace()
            main = fluid.Program()
            with fluid.program_guard(main):
                x = fluid.layers.data(name='x', shape=[13], dtype='float32')
                y = fluid.layers.data(name='y', shape=[1], dtype='float32')
                y_predict = fluid.layers.fc(input=x, size=1, act=None)
                cost = fluid.layers.square_error_cost(input=y_predict, label=y)
                avg_cost = fluid.layers.mean(cost)

                moment_optimizer = fluid.optimizer.MomentumOptimizer(learning_rate=0.001, momentum=0.9)
                moment_optimizer.minimize(avg_cost)

                fetch_list = [avg_cost]
                train_reader = paddle.batch(
                    paddle.dataset.uci_housing.train(), batch_size=1)
                feeder = fluid.DataFeeder(place=place, feed_list=[x, y])
                exe = fluid.Executor(place)
                exe.run(fluid.default_startup_program())
                for data in train_reader():
                    exe.run(main, feed=feeder.feed(data), fetch_list=fetch_list)

957 958 959
    """
    _velocity_acc_str = "velocity"

X
Xin Pan 已提交
960 961 962
    def __init__(self,
                 learning_rate,
                 momentum,
963
                 parameter_list=None,
X
Xin Pan 已提交
964 965 966
                 use_nesterov=False,
                 regularization=None,
                 name=None):
967 968
        assert learning_rate is not None
        assert momentum is not None
Q
Qiao Longfei 已提交
969
        super(MomentumOptimizer, self).__init__(
X
Xin Pan 已提交
970
            learning_rate=learning_rate,
971
            parameter_list=parameter_list,
X
Xin Pan 已提交
972 973
            regularization=regularization,
            name=name)
974 975
        self.type = "momentum"
        self._momentum = momentum
976
        self._use_nesterov = bool(use_nesterov)
977 978 979 980 981

    def _create_accumulators(self, block, parameters):
        assert isinstance(block, framework.Block)

        for p in parameters:
Q
Qiao Longfei 已提交
982
            self._add_accumulator(self._velocity_acc_str, p)
983 984 985 986 987 988

    def _append_optimize_op(self, block, param_and_grad):
        assert isinstance(block, framework.Block)

        velocity_acc = self._get_accumulator(self._velocity_acc_str,
                                             param_and_grad[0])
989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006
        attrs = {"mu": self._momentum, "use_nesterov": self._use_nesterov}

        inputs = {
            "Param": [param_and_grad[0]],
            "Grad": [param_and_grad[1]],
            "Velocity": [velocity_acc],
            "LearningRate": [self._create_param_lr(param_and_grad)]
        }

        outputs = {
            "ParamOut": [param_and_grad[0]],
            "VelocityOut": [velocity_acc]
        }

        if framework.in_dygraph_mode():
            core.ops.momentum(inputs, attrs, outputs)
            return None

1007 1008 1009
        # create the momentum optimize op
        momentum_op = block.append_op(
            type=self.type,
1010 1011 1012
            inputs=inputs,
            outputs=outputs,
            attrs=attrs,
M
minqiyang 已提交
1013
            stop_gradient=True)
1014 1015

        return momentum_op
1016 1017


1018
class DGCMomentumOptimizer(Optimizer):
1019
    """
1020
    DGC (Deep Gradient Compression) Momentum Optimizer. Original paper is https://arxiv.org/abs/1712.01887
1021

G
gongweibao 已提交
1022
    DGC reduces the communication bandwidth by sending only the important gradients (sparse update):\
1023 1024
        only gradients larger than a threshold are transmitted.

G
gongweibao 已提交
1025
    To avoid losing information, DGC accumulates the rest of the gradients locally.
1026 1027 1028

    Eventually, these gradients become large enough to be transmitted.

1029
    Thus, DGC sends the large gradients immediately but eventually sends all of the gradients over time.
1030

G
gongweibao 已提交
1031
    To ensure no loss of accuracy, DGC employs momentum correction and local gradient clipping on top of the gradient sparsification to maintain model performance.
1032 1033 1034 1035

    DGC also uses momentum factor masking and warmup training to overcome the staleness problem caused by reduced communication.

    This optimizer will do two things:
1036

1037 1038
        1. Compress the gradient by get TopK import value from tensor \
            and use it for allreduce to reduce network bandwidth.
1039

1040
        2. Call momentum to optimize the cost.
1041 1042

    Args:
1043 1044
        learning_rate (float|Variable): The learning rate used to update parameters. \
            It can be a float value or a Variable with one float value as a data element.
1045
        momentum (float): Momentum factor.
G
gongweibao 已提交
1046
        rampup_begin_step (int): The beginning step from which gradient compression is implemented.
1047 1048 1049 1050 1051 1052 1053
        rampup_step (int): Time steps used in sparsity warm-up periods. Default is 1.
            For example, if the sparsity is [0.75, 0.9375, 0.984375, 0.996, 0.999], and the rampup_step is 100, \
                it will use 0.75 at 0~19 steps, and 0.9375 at 20~39 steps, and so on. \
                And when reach sparsity array ends, it will use 0.999 then and after.
        sparsity (list[float]): Get top important element from gradient tensor, the ratio is (1 - current sparsity). \
            Default is [0.999]. For example, if the sparsity is [0.99, 0.999], \
                the top [1%, 0.1%] important element will be transmitted.
1054 1055 1056
        parameter_list (list, optional):  List of ``Variable`` names to update to minimize ``loss``. \
            This parameter is required in dygraph mode. \
            The default value is None in static mode, at this time all parameters will be updated.
1057 1058 1059 1060 1061 1062 1063
        use_nesterov (bool): Enables Nesterov momentum. True means use Nesterov. Default is False.
        local_grad_clip_norm (float, optional): Local gradient clip norm value. Optional, default is None, represent no need clip.
        num_trainers (int, optional): The number of training nodes. Optional, default is None.
        regularization (WeightDecayRegularizer, optional): A Regularizer, such as \
            :ref:`api_fluid_regularizer_L2DecayRegularizer`. Optional, default is None.
        name (str, optional): This parameter is used by developers to print debugging information. \
            For details, please refer to :ref:`api_guide_Name`. Default is None.
1064 1065 1066 1067

    Examples:
        .. code-block:: python

1068
            import paddle.fluid as fluid
1069
            optimizer = fluid.optimizer.DGCMomentumOptimizer(
G
gongweibao 已提交
1070 1071 1072 1073 1074
                        learning_rate=0.0001,
                        momentum=0.9,
                        rampup_step=1000,
                        rampup_begin_step=1252,
                        sparsity=[0.999, 0.999])
1075 1076

    """
1077 1078
    _u_velocity_acc_str = "_dgc_u_"
    _v_velocity_acc_str = "_dgc_v_"
1079 1080 1081 1082 1083 1084 1085

    def __init__(self,
                 learning_rate,
                 momentum,
                 rampup_begin_step,
                 rampup_step=1,
                 sparsity=[0.999],
1086
                 parameter_list=None,
1087 1088 1089 1090 1091
                 use_nesterov=False,
                 local_grad_clip_norm=None,
                 num_trainers=None,
                 regularization=None,
                 name=None):
Z
zhongpu 已提交
1092 1093
        if framework.in_dygraph_mode():
            raise Exception("In dygraph, don't support DGCMomentumOptimizer.")
1094 1095 1096 1097
        assert learning_rate is not None
        assert momentum is not None
        super(DGCMomentumOptimizer, self).__init__(
            learning_rate=learning_rate,
1098
            parameter_list=parameter_list,
1099 1100 1101 1102 1103
            regularization=regularization,
            name=name)
        self.type = "dgc_momentum"
        self._momentum = momentum
        self._use_nesterov = bool(use_nesterov)
1104

1105
        assert rampup_begin_step >= 0, "rampup_begin_step must >= 0"
1106
        self._rampup_begin_step = rampup_begin_step
1107 1108
        self._rampup_step = rampup_step
        self._sparsity = sparsity
1109

1110
        self._rampup_begin_step_var = None
1111
        self._global_step_var = None
1112

1113 1114 1115 1116 1117 1118 1119 1120 1121
        self._local_grad_clip_norm = None
        self._clip_norm = None
        if local_grad_clip_norm is not None:
            assert isinstance(num_trainers, int)
            assert isinstance(local_grad_clip_norm, float)
            assert num_trainers > 0

            self._local_grad_clip_norm = local_grad_clip_norm
            self._num_trainers = num_trainers
1122
            self._clip_norm = local_grad_clip_norm * (num_trainers**-0.5)
1123

1124 1125
        self.regular_type, self.regular_coeff = self._get_regularization_param(
            self.regularization)
1126

1127 1128 1129
    def _get_regularization_param(self, regularization):
        regular_type = 0
        regular_coeff = 0.0
1130

1131 1132
        if regularization is not None:
            regular_coeff = regularization._regularization_coeff
1133
            from .regularizer import L1Decay, L2Decay
1134 1135 1136 1137
            if isinstance(regularization, L1Decay):
                regular_type = 1
            elif isinstance(regularization, L2Decay):
                regular_type = 2
1138 1139
            else:
                assert False, 'regularization must be None|L1Decay|L2Deacy'
1140
        return regular_type, regular_coeff
1141

1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152
    def _is_use_dgc(self, param_var, grad_var):
        var_numel = abs(reduce(lambda x, y: x * y, param_var.shape))
        if var_numel < 16384 or \
           param_var.type == core.VarDesc.VarType.SELECTED_ROWS  or \
           grad_var.type == core.VarDesc.VarType.SELECTED_ROWS  or  \
               param_var.dtype != core.VarDesc.VarType.FP32 :
            return False
        return True

    def _append_optimize_op(self, block, param_and_grad):
        assert isinstance(block, framework.Block)
1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167
        velocity_acc = self._get_accumulator(self._u_velocity_acc_str,
                                             param_and_grad[0])
        assert velocity_acc is not None

        inputs = {
            "Param": param_and_grad[0],
            "Grad": param_and_grad[1],
            "Velocity": velocity_acc,
            "LearningRate": self._create_param_lr(param_and_grad),
        }
        outputs = {
            "ParamOut": param_and_grad[0],
            "VelocityOut": velocity_acc,
        }
        attrs = {"mu": self._momentum, "use_nesterov": self._use_nesterov}
1168 1169

        if not self._is_use_dgc(param_and_grad[0], param_and_grad[1]):
1170 1171 1172
            type = "momentum"
        else:
            type = "dgc_momentum"
1173 1174 1175 1176 1177
            inputs.update({
                "current_step": self._global_step_var,
                "nranks": self._nranks_var
            })
            outputs.update({'Grad_out': param_and_grad[1]})
1178
            attrs.update({"rampup_begin_step": float(self._rampup_begin_step)})
1179 1180 1181

        # create the dgc momentum optimize op
        dgc_momentum_op = block.append_op(
1182 1183 1184 1185
            type=type,
            inputs=inputs,
            outputs=outputs,
            attrs=attrs,
1186 1187 1188
            stop_gradient=True)
        return dgc_momentum_op

1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207
    def _add_auto_increment_var(self, counter_name, begin, step=1):
        helper = LayerHelper('global_step_counter')
        counter, is_new_var = helper.create_or_get_global_variable(
            name=counter_name, dtype='float32', shape=[1], persistable=True)
        if is_new_var:
            helper.set_variable_initializer(
                counter,
                initializer=Constant(
                    value=float(begin - 1), force_cpu=True))
            helper.main_program.global_block()._prepend_op(
                type='increment',
                inputs={'X': [counter]},
                outputs={'Out': [counter]},
                attrs={'step': float(step)},
                stop_gradient=True)
            counter.stop_gradient = True

        return counter

1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220
    def _add_nranks_var(self, name, value=-1):
        helper = LayerHelper('global_step_counter')
        counter, is_new_var = helper.create_or_get_global_variable(
            name=name, dtype='float32', shape=[1], persistable=True)
        if is_new_var:
            helper.set_variable_initializer(
                counter,
                initializer=Constant(
                    value=float(value), force_cpu=True))
            counter.stop_gradient = True

        return counter

1221 1222 1223 1224 1225 1226
    def _append_dgc_ops(self, param_and_grads):
        main_program = default_main_program()
        main_program._enable_dgc = True

        # step counter
        self._global_step_var = self._add_auto_increment_var(
G
gongweibao 已提交
1227
            counter_name=core.dgc.kDGCCounterName(), begin=0)
1228

1229 1230 1231
        self._nranks_var = self._add_nranks_var(
            name=core.dgc.kDGCNRanksName(), value=-1)

1232 1233 1234 1235 1236
        # rampup begin step var for all_reduce_op_handle
        self._rampup_begin_step_var = tensor.create_global_var(
            shape=[1],
            dtype=core.VarDesc.VarType.FP32,
            persistable=True,
G
gongweibao 已提交
1237
            name=core.dgc.kDGCRampUpBeginStepName(),
1238 1239 1240
            value=self._rampup_begin_step * 1.0,
            force_cpu=True)

1241 1242
        self.helper = LayerHelper(self.__class__.__name__)

1243
        for param_var, grad_var in param_and_grads:
1244 1245 1246
            # reuse velocity in dgc_op and dgc_momentum_op
            u_var = self._add_accumulator(self._u_velocity_acc_str, param_var)

1247
            if not self._is_use_dgc(param_var, grad_var):
1248 1249
                continue

1250
            v_var = self._add_accumulator(self._v_velocity_acc_str, param_var)
1251 1252 1253 1254 1255

            k_var = tensor.create_global_var(
                shape=[1],
                dtype=param_var.dtype,
                persistable=True,
G
gongweibao 已提交
1256
                name=param_var.name + core.dgc.kDGCKName(),
1257 1258 1259 1260 1261 1262 1263
                value=0.0,
                force_cpu=True)

            encoded_var = tensor.create_global_var(
                shape=[1],
                dtype=param_var.dtype,
                persistable=True,
G
gongweibao 已提交
1264
                name=param_var.name + core.dgc.kDGCEncodedName(),
1265 1266 1267
                value=0.0,
                force_cpu=False)

1268 1269 1270 1271 1272 1273 1274 1275
            gather_var = tensor.create_global_var(
                shape=[1],
                dtype=param_var.dtype,
                persistable=True,
                name=param_var.name + core.dgc.kDGCGatherName(),
                value=0.0,
                force_cpu=False)

1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297
            # del back oprolevarname
            op_maker = core.op_proto_and_checker_maker
            backward = core.op_proto_and_checker_maker.OpRole.Backward
            for op in main_program.global_block().ops:
                if not self._is_the_backward_op(op):
                    continue

                var_attr = op.all_attrs()[op_maker.kOpRoleVarAttrName()]
                if param_var.name not in var_attr:
                    continue

                var_attr.remove(param_var.name)
                var_attr.remove(grad_var.name)
                if len(var_attr) > 1:
                    op._set_attr(op_maker.kOpRoleVarAttrName(), var_attr)
                else:
                    op._remove_attr(op_maker.kOpRoleVarAttrName())

            clip_var = grad_var
            if self._local_grad_clip_norm is not None:
                clip_var = self._append_clip_norm(grad_var, self._clip_norm)
            self._dgc_op(param_var, clip_var, grad_var, u_var, v_var, k_var,
1298
                         encoded_var, gather_var)
1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313

    def _is_the_backward_op(self, op):
        op_maker = core.op_proto_and_checker_maker
        backward = core.op_proto_and_checker_maker.OpRole.Backward
        if op_maker.kOpRoleVarAttrName() in op.attr_names and \
                int(op.all_attrs()[op_maker.kOpRoleAttrName()]) == int(backward):
            return True
        return False

    def _clip_by_norm(self, x, max_norm, name=None):
        args = {'x': x, 'max_norm': max_norm, 'name': name}

        helper = LayerHelper("dgc_clip_by_norm_op", **args)

        if name is None:
1314 1315
            name = unique_name.generate_with_ignorable_key(".".join(
                [helper.name, 'tmp']))
1316 1317 1318 1319 1320

        out = helper.create_variable(
            type=x.type, name=name, dtype=x.dtype, persistable=False)

        helper.append_op(
G
gongweibao 已提交
1321
            type="dgc_clip_by_norm",
1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333
            inputs={"X": x,
                    "current_step": self._global_step_var},
            attrs={
                "max_norm": max_norm,
                "rampup_begin_step": float(self._rampup_begin_step)
            },
            outputs={"Out": out})
        return out

    def _append_clip_norm(self, grad_var, clip_norm):
        with grad_var.block.program._backward_role_guard():
            return self._clip_by_norm(
G
gongweibao 已提交
1334
                x=grad_var, max_norm=clip_norm, name=grad_var.name)
1335 1336

    def _dgc_op(self, param_var, clip_var, grad_var, u_var, v_var, k_var,
1337
                encoded_var, gather_var):
1338 1339
        block = framework.default_main_program().global_block()
        op_maker = core.op_proto_and_checker_maker
1340

1341 1342 1343 1344 1345 1346 1347
        regular_type = self.regular_type
        regular_coeff = self.regular_coeff
        # The regularizer of the Parameters have higher priority
        if param_var.regularizer is not None:
            regular_type, regular_coeff = self._get_regularization_param(
                param_var.regularizer)

1348 1349 1350 1351 1352 1353
        dgc_op = block.append_op(
            type="dgc",
            inputs={
                "U": u_var,
                "V": v_var,
                "Grad": clip_var,
1354
                "Param": param_var,
1355 1356
                "current_step": self._global_step_var,
                "nranks": self._nranks_var,
1357 1358 1359 1360 1361 1362
            },
            outputs={
                "U_out": u_var,
                "V_out": v_var,
                "EncodeGrad": encoded_var,
                "k": k_var,
1363 1364
                "Grad_out": grad_var,
                "GatherBuff": gather_var,
1365 1366 1367 1368 1369 1370
            },
            attrs={
                "m": self._momentum,
                "sparsity": self._sparsity,
                "use_nesterov": self._use_nesterov,
                "rampup_begin_step": float(self._rampup_begin_step),
1371
                "rampup_step": float(self._rampup_step),
1372 1373
                "regular_coeff": float(regular_coeff),
                "regular_type": int(regular_type),
1374 1375 1376 1377 1378 1379 1380 1381
            },
            stop_gradient=True)

        backward = op_maker.OpRole.Backward
        dgc_op._set_attr(op_maker.kOpRoleAttrName(), backward)
        dgc_op._set_attr(op_maker.kOpRoleVarAttrName(),
                         [param_var.name, grad_var.name])

1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412
    def apply_gradients(self, params_grads):
        params_grads = sorted(params_grads, key=lambda x: x[0].name)

        params_grads, table_param_and_grad, table_optimize_op = \
            self._process_distribute_lookuptable(params_grads)

        not_dgc_params_grads = []
        dgc_params_grads = []
        for param, grad in params_grads:
            if not self._is_use_dgc(param, grad):
                not_dgc_params_grads.append((param, grad))
            else:
                dgc_params_grads.append((param, grad))

        # DGC clip and regularization in local
        not_dgc_params_grads = append_gradient_clip_ops(not_dgc_params_grads)

        # Add regularization if any
        not_dgc_params_grads = append_regularization_ops(not_dgc_params_grads,
                                                         self.regularization)

        params_grads = not_dgc_params_grads + dgc_params_grads
        params_grads = sorted(params_grads, key=lambda x: x[0].name)

        optimize_ops = self._create_optimization_pass(params_grads)
        if table_optimize_op is not None:
            optimize_ops.append(table_optimize_op)
            params_grads.append(table_param_and_grad)

        return optimize_ops

1413

1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428
class LarsMomentumOptimizer(Optimizer):
    """
    Momentum optimizer with LARS support

    The update equations are as follows:

    .. math::

        & local\_learning\_rate = learning\_rate * lars\_coeff * \\
          \\frac{||param||}{||gradient|| + lars\_weight\_decay * ||param||}

        & velocity = mu * velocity + local\_learning\_rate * (gradient + lars\_weight\_decay * param)

        & param = param - velocity

1429 1430 1431 1432 1433 1434
    Parameters:
        learning_rate (float|Variable): The learning rate used to update parameters. \
            Can be a float value or a Variable with one float value as data element. \
            momentum (float): momentum factor
        lars_coeff (float): Defines how much we trust the layer to change its weights.
        lars_weight_decay (float): Weight decay coefficient for decaying using LARS.
1435 1436 1437
        parameter_list (list, optional):  List of ``Variable`` names to update to minimize ``loss``. \
            This parameter is required in dygraph mode. \
            The default value is None in static mode, at this time all parameters will be updated.
1438 1439 1440 1441
        regularization: A Regularizer, such as :ref:`api_fluid_regularizer_L2DecayRegularizer`.
            Optional, default is None.
        name (str, optional): This parameter is used by developers to print debugging information. \
            For details, please refer to :ref:`api_guide_Name`. Default is None.
1442 1443 1444 1445

    Examples:
        .. code-block:: python

1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461
            import paddle.fluid as fluid
            import numpy as np

            np_inp = np.array([[1.0, 2.0], [3.0, 4.0]], dtype=np.float32)
            inp = fluid.layers.data(
                name="inp", shape=[2, 2], append_batch_size=False)
            out = fluid.layers.fc(inp, size=3)
            out = fluid.layers.reduce_sum(out)
            optimizer = fluid.optimizer.LarsMomentumOptimizer(learning_rate=0.001, momentum=0.9)
            optimizer.minimize(out)

            exe = fluid.Executor(fluid.CPUPlace())
            exe.run(fluid.default_startup_program())
            exe.run(
                feed={"inp": np_inp},
                fetch_list=[out.name])
1462 1463 1464 1465 1466 1467 1468 1469
    """
    _velocity_acc_str = "velocity"

    def __init__(self,
                 learning_rate,
                 momentum,
                 lars_coeff=0.001,
                 lars_weight_decay=0.0005,
1470
                 parameter_list=None,
1471 1472 1473 1474 1475 1476
                 regularization=None,
                 name=None):
        assert learning_rate is not None
        assert momentum is not None
        super(LarsMomentumOptimizer, self).__init__(
            learning_rate=learning_rate,
1477
            parameter_list=parameter_list,
1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512
            regularization=regularization,
            name=name)
        self.type = "lars_momentum"
        self._momentum = momentum
        self._lars_coeff = float(lars_coeff)
        self._lars_weight_decay = float(lars_weight_decay)

    def _create_accumulators(self, block, parameters):
        assert isinstance(block, framework.Block)

        for p in parameters:
            self._add_accumulator(self._velocity_acc_str, p)

    def _append_optimize_op(self, block, param_and_grad):
        assert isinstance(block, framework.Block)

        velocity_acc = self._get_accumulator(self._velocity_acc_str,
                                             param_and_grad[0])
        # create the momentum optimize op
        momentum_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
                "Velocity": velocity_acc,
                "LearningRate": self._create_param_lr(param_and_grad)
            },
            outputs={
                "ParamOut": param_and_grad[0],
                "VelocityOut": velocity_acc
            },
            attrs={
                "mu": self._momentum,
                "lars_coeff": self._lars_coeff,
                "lars_weight_decay": self._lars_weight_decay
M
minqiyang 已提交
1513 1514
            },
            stop_gradient=True)
1515 1516 1517 1518

        return momentum_op


1519
class AdagradOptimizer(Optimizer):
Q
qiaolongfei 已提交
1520
    """
1521 1522
    The Adaptive Gradient optimizer (Adagrad for short) can adaptively assign
    different learning rates to individual parameters.
Q
qiaolongfei 已提交
1523

1524
    The parameter ``param_out`` update rule with gradient ``grad``:
Q
qiaolongfei 已提交
1525 1526 1527 1528 1529 1530 1531

    .. math::

        moment\_out &= moment + grad * grad

        param\_out &= param - \\frac{learning\_rate * grad}{\sqrt{moment\_out} + \epsilon}

1532 1533 1534 1535 1536 1537
    Related paper: `Adaptive Subgradient Methods for Online Learning and
    Stochastic Optimization <http://www.jmlr.org/papers/volume12/duchi11a/duchi11a.pdf>`_.

    The original paper does not have the ``epsilon`` attribute. It is added here
    in our implementation as also proposed `Per-parameter adaptive learning rate
    methods <http://cs231n.github.io/neural-networks-3/#ada>`_
Q
qiaolongfei 已提交
1538 1539 1540
    for numerical stability to avoid the division by zero error.

    Args:
1541 1542 1543 1544
        learning_rate (float|Variable): The learning rate used to update ``Parameter``.
            It can be a float value or a ``Variable`` with a float type.
        epsilon (float, optional): A small float value for numerical stability.
            The default value is 1e-06.
1545 1546 1547
        parameter_list (list, optional):  List of ``Variable`` names to update to minimize ``loss``. \
            This parameter is required in dygraph mode. \
            The default value is None in static mode, at this time all parameters will be updated.
1548 1549 1550 1551 1552 1553 1554
        regularization (WeightDecayRegularizer, optional): A ``Regularizer``, such as
             :ref:`api_fluid_regularizer_L2DecayRegularizer`. The default value is None.
        name (str, optional): Normally there is no need for user to set this property.
            For more information, please refer to :ref:`api_guide_Name`.
            The default value is None.
        initial_accumulator_value (float, optional): Initial value for moment accumulator.
            The default value is 0.0.
Q
qiaolongfei 已提交
1555 1556 1557 1558

    Examples:
        .. code-block:: python

1559
            import numpy as np
1560
            import paddle.fluid as fluid
1561 1562

            np_inp = np.array([[1.0, 2.0], [3.0, 4.0]], dtype=np.float32)
1563
            inp = fluid.data(name="inp", shape=[2, 2])
1564 1565
            out = fluid.layers.fc(inp, size=3)
            out = fluid.layers.reduce_sum(out)
1566
            optimizer = fluid.optimizer.AdagradOptimizer(learning_rate=0.2)
1567 1568 1569 1570 1571 1572 1573
            optimizer.minimize(out)

            exe = fluid.Executor(fluid.CPUPlace())
            exe.run(fluid.default_startup_program())
            exe.run(
                feed={"inp": np_inp},
                fetch_list=[out.name])
1574 1575 1576
    """
    _moment_acc_str = "moment"

X
Xin Pan 已提交
1577 1578 1579
    def __init__(self,
                 learning_rate,
                 epsilon=1.0e-6,
1580
                 parameter_list=None,
X
Xin Pan 已提交
1581
                 regularization=None,
1582
                 name=None,
X
xuezhong 已提交
1583
                 initial_accumulator_value=0.0):
1584 1585
        assert learning_rate is not None
        assert epsilon is not None
Q
Qiao Longfei 已提交
1586
        super(AdagradOptimizer, self).__init__(
X
Xin Pan 已提交
1587
            learning_rate=learning_rate,
1588
            parameter_list=parameter_list,
X
Xin Pan 已提交
1589 1590
            regularization=regularization,
            name=name)
1591 1592
        self.type = "adagrad"
        self._epsilon = epsilon
1593
        self.initial_accumulator_value = initial_accumulator_value
1594 1595 1596 1597 1598

    def _create_accumulators(self, block, parameters):
        assert isinstance(block, framework.Block)

        for p in parameters:
Z
zhongpu 已提交
1599 1600 1601 1602
            self._add_accumulator(
                self._moment_acc_str,
                p,
                fill_value=self.initial_accumulator_value)
1603 1604 1605 1606 1607 1608

    def _append_optimize_op(self, block, param_and_grad):
        assert isinstance(block, framework.Block)

        moment_acc = self._get_accumulator(self._moment_acc_str,
                                           param_and_grad[0])
1609
        # Create the adagrad optimizer op
1610 1611 1612 1613 1614 1615
        adagrad_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
                "Moment": moment_acc,
1616
                "LearningRate": self._create_param_lr(param_and_grad)
1617 1618 1619
            },
            outputs={"ParamOut": param_and_grad[0],
                     "MomentOut": moment_acc},
M
minqiyang 已提交
1620 1621
            attrs={"epsilon": self._epsilon},
            stop_gradient=True)
1622 1623

        return adagrad_op
1624 1625 1626


class AdamOptimizer(Optimizer):
Q
qiaolongfei 已提交
1627
    """
T
tianshuo78520a 已提交
1628
    The Adam optimizer uses an optimization described at the end
1629 1630 1631 1632 1633
    of section 2 of `Adam paper <https://arxiv.org/abs/1412.6980>`_ ,
    it can dynamically adjusts the learning rate of each parameter using
    the 1st moment estimates and the 2nd moment estimates of the gradient.
    
    The parameter ``param_out`` update rule with gradient ``grad``:
Q
qiaolongfei 已提交
1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647

    .. math::

        t & = t + 1

        moment\_1\_out & = {\\beta}_1 * moment\_1 + (1 - {\\beta}_1) * grad

        moment\_2\_out & = {\\beta}_2 * moment\_2 + (1 - {\\beta}_2) * grad * grad

        learning\_rate & = learning\_rate * \\
                          \\frac{\sqrt{1 - {\\beta}_2^t}}{1 - {\\beta}_1^t}

        param\_out & = param - learning\_rate * \\frac{moment\_1}{\sqrt{moment\_2} + \epsilon}

1648 1649
    Related paper: `Adam: A Method for Stochastic Optimization <https://arxiv.org/abs/1412.6980>`_

Q
qiaolongfei 已提交
1650
    Args:
1651 1652
        learning_rate (float|Variable, optional): The learning rate used to update ``Parameter``.
            It can be a float value or a ``Variable`` with a float type. The default value is 0.001.
1653 1654
        beta1 (float|Variable, optional): The exponential decay rate for the 1st moment estimates.
            It should be a float number or a Variable with shape [1] and data type as float32.
1655
            The default value is 0.9.
1656 1657
        beta2 (float|Variable, optional): The exponential decay rate for the 2nd moment estimates.
            It should be a float number or a Variable with shape [1] and data type as float32.
1658 1659 1660
            The default value is 0.999.
        epsilon (float, optional): A small float value for numerical stability.
            The default value is 1e-08.
1661 1662 1663
        parameter_list (list, optional):  List of ``Variable`` names to update to minimize ``loss``. \
            This parameter is required in dygraph mode. \
            The default value is None in static mode, at this time all parameters will be updated.
1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675
        regularization (WeightDecayRegularizer, optional): A ``Regularizer``, such as
             :ref:`api_fluid_regularizer_L2DecayRegularizer`. The default value is None.
        name (str, optional): Normally there is no need for user to set this property.
            For more information, please refer to :ref:`api_guide_Name`.
            The default value is None.
        lazy_mode (bool, optional): The official Adam algorithm has two moving-average accumulators.
            The accumulators are updated at every step. Every element of the two moving-average
            is updated in both dense mode and sparse mode. If the size of parameter is very large,
            then the update may be very slow. The lazy mode only update the element that has
            gradient in current mini-batch, so it will be much more faster. But this mode has
            different semantics with the original Adam algorithm and may lead to different result.
            The default value is False.
Q
qiaolongfei 已提交
1676 1677 1678 1679

    Examples:
        .. code-block:: python

1680 1681 1682 1683 1684 1685
            import paddle
            import paddle.fluid as fluid

            place = fluid.CPUPlace()
            main = fluid.Program()
            with fluid.program_guard(main):
1686 1687
                x = fluid.data(name='x', shape=[None, 13], dtype='float32')
                y = fluid.data(name='y', shape=[None, 1], dtype='float32')
1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702
                y_predict = fluid.layers.fc(input=x, size=1, act=None)
                cost = fluid.layers.square_error_cost(input=y_predict, label=y)
                avg_cost = fluid.layers.mean(cost)

                adam_optimizer = fluid.optimizer.AdamOptimizer(0.01)
                adam_optimizer.minimize(avg_cost)

                fetch_list = [avg_cost]
                train_reader = paddle.batch(
                    paddle.dataset.uci_housing.train(), batch_size=1)
                feeder = fluid.DataFeeder(place=place, feed_list=[x, y])
                exe = fluid.Executor(place)
                exe.run(fluid.default_startup_program())
                for data in train_reader():
                    exe.run(main, feed=feeder.feed(data), fetch_list=fetch_list)
Q
qiaolongfei 已提交
1703

1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720
        .. code-block:: python

            # Adam with beta1/beta2 as Variable
            import paddle
            import paddle.fluid as fluid
            import paddle.fluid.layers.learning_rate_scheduler as lr_scheduler

            place = fluid.CPUPlace()
            main = fluid.Program()
            with fluid.program_guard(main):
                x = fluid.data(name='x', shape=[None, 13], dtype='float32')
                y = fluid.data(name='y', shape=[None, 1], dtype='float32')
                y_predict = fluid.layers.fc(input=x, size=1, act=None)
                cost = fluid.layers.square_error_cost(input=y_predict, label=y)
                avg_cost = fluid.layers.mean(cost)

                # define beta decay variable
1721
                def get_decayed_betas(beta1_init, beta2_init, decay_steps, decay_rate):
1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749
                    global_step = lr_scheduler._decay_step_counter()

                    beta1 = fluid.layers.create_global_var(
                        shape=[1],
                        value=float(beta1_init),
                        dtype='float32',
                        # set persistable for save checkpoints and resume
                        persistable=True,
                        name="beta1")
                    beta2 = fluid.layers.create_global_var(
                        shape=[1],
                        value=float(beta2_init),
                        dtype='float32',
                        # set persistable for save checkpoints and resume
                        persistable=True,
                        name="beta2")

                    div_res = global_step / decay_steps
                    decayed_beta1 = beta1_init * (decay_rate**div_res)
                    decayed_beta2 = beta2_init * (decay_rate**div_res)
                    fluid.layers.assign(decayed_beta1, beta1)
                    fluid.layers.assign(decayed_beta2, beta2)

                    return beta1, beta2

                beta1, beta2 = get_decayed_betas(0.9, 0.99, 1e5, 0.9)
                adam_optimizer = fluid.optimizer.AdamOptimizer(
                                                    learning_rate=0.01,
1750
                                                    beta1=beta1,
1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761
                                                    beta2=beta2)
                adam_optimizer.minimize(avg_cost)

                fetch_list = [avg_cost]
                train_reader = paddle.batch(
                    paddle.dataset.uci_housing.train(), batch_size=1)
                feeder = fluid.DataFeeder(place=place, feed_list=[x, y])
                exe = fluid.Executor(place)
                exe.run(fluid.default_startup_program())
                for data in train_reader():
                    exe.run(main, feed=feeder.feed(data), fetch_list=fetch_list)
1762 1763 1764
    """
    _moment1_acc_str = "moment1"
    _moment2_acc_str = "moment2"
Q
qiaolongfei 已提交
1765 1766
    _beta1_pow_acc_str = "beta1_pow_acc"
    _beta2_pow_acc_str = "beta2_pow_acc"
1767 1768 1769 1770 1771

    def __init__(self,
                 learning_rate=0.001,
                 beta1=0.9,
                 beta2=0.999,
1772
                 epsilon=1e-8,
1773
                 parameter_list=None,
X
Xin Pan 已提交
1774
                 regularization=None,
Q
Qiao Longfei 已提交
1775
                 name=None,
Q
Qiao Longfei 已提交
1776
                 lazy_mode=False):
1777 1778 1779 1780
        assert learning_rate is not None
        assert beta1 is not None
        assert beta2 is not None
        assert epsilon is not None
Q
Qiao Longfei 已提交
1781
        super(AdamOptimizer, self).__init__(
X
Xin Pan 已提交
1782
            learning_rate=learning_rate,
1783
            parameter_list=parameter_list,
X
Xin Pan 已提交
1784 1785
            regularization=regularization,
            name=name)
1786 1787 1788 1789
        self.type = "adam"
        self._beta1 = beta1
        self._beta2 = beta2
        self._epsilon = epsilon
Q
Qiao Longfei 已提交
1790
        self._lazy_mode = lazy_mode
1791 1792 1793 1794 1795 1796

    def _create_accumulators(self, block, parameters):
        assert isinstance(block, framework.Block)

        # Create accumulator tensors for first and second moments
        for p in parameters:
Q
Qiao Longfei 已提交
1797 1798
            self._add_accumulator(self._moment1_acc_str, p)
            self._add_accumulator(self._moment2_acc_str, p)
Q
qiaolongfei 已提交
1799 1800 1801
            self._add_accumulator(
                name=self._beta1_pow_acc_str,
                param=p,
1802 1803
                fill_value=0.9 if isinstance(self._beta1, Variable) \
                        else self._beta1,
1804 1805
                shape=[1],
                type=core.VarDesc.VarType.LOD_TENSOR)
Q
qiaolongfei 已提交
1806 1807 1808
            self._add_accumulator(
                name=self._beta2_pow_acc_str,
                param=p,
1809 1810
                fill_value=0.999 if isinstance(self._beta2, Variable) \
                        else self._beta2,
1811 1812
                shape=[1],
                type=core.VarDesc.VarType.LOD_TENSOR)
1813 1814 1815 1816 1817 1818 1819 1820

    def _append_optimize_op(self, block, param_and_grad):
        assert isinstance(block, framework.Block)

        moment1 = self._get_accumulator(self._moment1_acc_str,
                                        param_and_grad[0])
        moment2 = self._get_accumulator(self._moment2_acc_str,
                                        param_and_grad[0])
Q
qiaolongfei 已提交
1821 1822 1823 1824 1825
        beta1_pow_acc = self._get_accumulator(self._beta1_pow_acc_str,
                                              param_and_grad[0])
        beta2_pow_acc = self._get_accumulator(self._beta2_pow_acc_str,
                                              param_and_grad[0])

1826
        # create the adam optimize op
1827
        inputs = {
1828 1829 1830 1831 1832 1833 1834
            "Param": [param_and_grad[0]],
            "Grad": [param_and_grad[1]],
            "LearningRate": [self._create_param_lr(param_and_grad)],
            "Moment1": [moment1],
            "Moment2": [moment2],
            "Beta1Pow": [beta1_pow_acc],
            "Beta2Pow": [beta2_pow_acc]
1835 1836
        }
        outputs = {
1837 1838 1839 1840 1841
            "ParamOut": [param_and_grad[0]],
            "Moment1Out": [moment1],
            "Moment2Out": [moment2],
            "Beta1PowOut": [beta1_pow_acc],
            "Beta2PowOut": [beta2_pow_acc],
1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857
        }
        attrs = {
            "epsilon": self._epsilon,
            "lazy_mode": self._lazy_mode,
            "min_row_size_to_use_multithread": 1000
        }

        if isinstance(self._beta1, Variable):
            inputs['Beta1Tensor'] = self._beta1
        else:
            attrs['beta1'] = self._beta1
        if isinstance(self._beta2, Variable):
            inputs['Beta2Tensor'] = self._beta2
        else:
            attrs['beta2'] = self._beta2

1858 1859 1860 1861
        if framework.in_dygraph_mode():
            core.ops.adam(inputs, attrs, outputs)
            return None

1862 1863
        adam_op = block.append_op(
            type=self.type,
1864 1865 1866
            inputs=inputs,
            outputs=outputs,
            attrs=attrs,
M
minqiyang 已提交
1867
            stop_gradient=True)
1868 1869 1870

        return adam_op

1871 1872

class AdamaxOptimizer(Optimizer):
Q
qiaolongfei 已提交
1873
    """
1874 1875 1876 1877
    The Adamax optimizer is implemented based on the Adamax Optimization 
    in Section 7 of `Adam paper <https://arxiv.org/abs/1412.6980>`_.
    The Adamax algorithm is a variant of the Adam algorithm based on the infinite norm,
    which makes the learning rate update algorithm more stable and simple.
Q
qiaolongfei 已提交
1878

1879
    The parameter ``param_out`` update rule with gradient ``grad``:
Q
qiaolongfei 已提交
1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892

    .. math::

        t & = t + 1

        moment\_out & = {\\beta}_1 * moment + (1 - {\\beta}_1) * grad

        inf\_norm\_out & = max({\\beta}_2 * inf\_norm + \epsilon, |grad|)

        learning\_rate & = \\frac{learning\_rate}{1 - {\\beta}_1^t}

        param\_out & = param - learning\_rate * \\frac{moment\_out}{inf\_norm\_out}

1893
    Related paper: `Adam: A Method for Stochastic Optimization <https://arxiv.org/abs/1412.6980>`_
Q
qiaolongfei 已提交
1894

1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906
    The original paper does not have an ``epsilon`` attribute,
    it is added here for numerical stability to prevent the division by 0 error.

    Args:
        learning_rate (float|Variable, optional): The learning rate used to update ``Parameter``.
            It can be a float value or a ``Variable`` with a float type. The default value is 0.001.
        beta1 (float, optional): The exponential decay rate for the 1st moment estimates.
            The default value is 0.9.
        beta2 (float, optional): The exponential decay rate for the 2nd moment estimates.
            The default value is 0.999.
        epsilon (float, optional): A small float value for numerical stability.
            The default value is 1e-08.
1907 1908 1909
        parameter_list (list, optional):  List of ``Variable`` names to update to minimize ``loss``. \
            This parameter is required in dygraph mode. \
            The default value is None in static mode, at this time all parameters will be updated.
1910 1911 1912 1913 1914 1915 1916 1917
        regularization (WeightDecayRegularizer, optional): A ``Regularizer``, such as
             :ref:`api_fluid_regularizer_L2DecayRegularizer`. The default value is None.
        name (str, optional): Normally there is no need for user to set this property.
            For more information, please refer to :ref:`api_guide_Name`.
            The default value is None.

    **Notes**:
        **Currently, AdamaxOptimizer doesn't support sparse parameter optimization.**
Q
qiaolongfei 已提交
1918

1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931
    Examples:
        .. code-block:: python

          import paddle.fluid as fluid
          import numpy

          # First create the Executor.
          place = fluid.CPUPlace() # fluid.CUDAPlace(0)
          exe = fluid.Executor(place)

          train_program = fluid.Program()
          startup_program = fluid.Program()
          with fluid.program_guard(train_program, startup_program):
1932
              data = fluid.data(name='X', shape=[None, 1], dtype='float32')
1933 1934
              hidden = fluid.layers.fc(input=data, size=10)
              loss = fluid.layers.mean(hidden)
1935
              adam = fluid.optimizer.AdamaxOptimizer(learning_rate=0.2)
1936 1937 1938 1939 1940 1941 1942 1943 1944
              adam.minimize(loss)

          # Run the startup program once and only once.
          exe.run(startup_program)

          x = numpy.random.random(size=(10, 1)).astype('float32')
          outs = exe.run(program=train_program,
                        feed={'X': x},
                         fetch_list=[loss.name])
1945 1946 1947
    """
    _moment_acc_str = "moment"
    _inf_norm_acc_str = "inf_norm"
Q
qiaolongfei 已提交
1948
    _beta1_pow_acc_str = "beta1_pow_acc"
1949 1950 1951 1952 1953

    def __init__(self,
                 learning_rate=0.001,
                 beta1=0.9,
                 beta2=0.999,
1954
                 epsilon=1e-8,
1955
                 parameter_list=None,
X
Xin Pan 已提交
1956 1957
                 regularization=None,
                 name=None):
1958 1959 1960 1961
        assert learning_rate is not None
        assert beta1 is not None
        assert beta2 is not None
        assert epsilon is not None
Q
Qiao Longfei 已提交
1962
        super(AdamaxOptimizer, self).__init__(
X
Xin Pan 已提交
1963
            learning_rate=learning_rate,
1964
            parameter_list=parameter_list,
X
Xin Pan 已提交
1965 1966
            regularization=regularization,
            name=name)
1967 1968 1969 1970 1971 1972 1973 1974
        self.type = "adamax"
        self._beta1 = beta1
        self._beta2 = beta2
        self._epsilon = epsilon

    def _create_accumulators(self, block, parameters):
        # Create accumulator tensors for first moment and infinity norm
        for p in parameters:
Q
Qiao Longfei 已提交
1975 1976
            self._add_accumulator(self._moment_acc_str, p)
            self._add_accumulator(self._inf_norm_acc_str, p)
Q
qiaolongfei 已提交
1977 1978 1979 1980 1981
            self._add_accumulator(
                name=self._beta1_pow_acc_str,
                param=p,
                fill_value=self._beta1,
                shape=[1])
1982 1983 1984 1985 1986 1987 1988

    def _append_optimize_op(self, block, param_and_grad):
        assert isinstance(block, framework.Block)

        moment = self._get_accumulator(self._moment_acc_str, param_and_grad[0])
        inf_norm = self._get_accumulator(self._inf_norm_acc_str,
                                         param_and_grad[0])
Q
qiaolongfei 已提交
1989 1990
        beta1_pow_acc = self._get_accumulator(self._beta1_pow_acc_str,
                                              param_and_grad[0])
1991 1992 1993 1994 1995 1996
        # create the adamax optimize op
        adamax_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
1997
                "LearningRate": self._create_param_lr(param_and_grad),
1998 1999
                "Moment": moment,
                "InfNorm": inf_norm,
Q
qiaolongfei 已提交
2000
                "Beta1Pow": beta1_pow_acc
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
            },
            outputs={
                "ParamOut": param_and_grad[0],
                "MomentOut": moment,
                "InfNormOut": inf_norm
            },
            attrs={
                "beta1": self._beta1,
                "beta2": self._beta2,
                "epsilon": self._epsilon
M
minqiyang 已提交
2011 2012
            },
            stop_gradient=True)
2013 2014 2015

        return adamax_op

2016
    def _finish_update(self, block, parameters_and_grads):
2017 2018 2019
        """Update Beta1 Power accumulator
        """
        assert isinstance(block, framework.Block)
2020
        for param, grad in parameters_and_grads:
C
chengduo 已提交
2021
            if grad is None or param.trainable is False:
2022
                continue
X
Xin Pan 已提交
2023 2024
            with param.block.program._optimized_guard(
                [param, grad]), name_scope('adamx'):
2025 2026
                beta1_pow_acc = self._get_accumulator(self._beta1_pow_acc_str,
                                                      param)
2027
                block.append_op(
2028 2029 2030
                    type="scale",
                    inputs={"X": beta1_pow_acc},
                    outputs={"Out": beta1_pow_acc},
M
minqiyang 已提交
2031 2032
                    attrs={"scale": self._beta1},
                    stop_gradient=True)
2033 2034


2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072
class DpsgdOptimizer(Optimizer):
    """
    We implement the Dpsgd optimizer according to CCS16 paper -
    Deep Learning with Differential Privacy.

    Examples:
        .. code-block:: python

          import paddle.fluid as fluid
          import numpy

          # First create the Executor.
          place = fluid.CPUPlace() # fluid.CUDAPlace(0)
          exe = fluid.Executor(place)

          train_program = fluid.Program()
          startup_program = fluid.Program()
          with fluid.program_guard(train_program, startup_program):
              data = fluid.layers.data(name='X', shape=[1], dtype='float32')
              hidden = fluid.layers.fc(input=data, size=10)
              loss = fluid.layers.mean(hidden)
              optimizer = fluid.optimizer.Dpsgd(learning_rate=0.01, clip=10.0, batch_size=16.0, sigma=1.0)
              optimizer.minimize(loss)

          # Run the startup program once and only once.
          exe.run(startup_program)

          x = numpy.random.random(size=(10, 1)).astype('float32')
          outs = exe.run(program=train_program,
                        feed={'X': x},
                         fetch_list=[loss.name])

    Args:
        learning_rate (float|Variable): the learning rate used to update parameters. \
        Can be a float value or a Variable with one float value as data element.
        clip (float): clipping threshold
        batch_size (float): batch size.
        sigma (float): for gaussian noise.
2073 2074 2075
        parameter_list (list, optional):  List of ``Variable`` names to update to minimize ``loss``. \
            This parameter is required in dygraph mode. \
            The default value is None in static mode, at this time all parameters will be updated.
2076 2077 2078 2079 2080 2081 2082 2083
    Notes:
       Currently, DpsgdOptimizer doesn't support sparse parameter optimization.
    """

    def __init__(self,
                 learning_rate=0.001,
                 clip=0.9,
                 batch_size=0.999,
2084 2085
                 sigma=1e-8,
                 parameter_list=None):
2086 2087 2088 2089
        assert learning_rate is not None
        assert clip is not None
        assert batch_size is not None
        assert sigma is not None
2090 2091
        super(DpsgdOptimizer, self).__init__(
            learning_rate=learning_rate, parameter_list=parameter_list)
2092 2093 2094 2095
        self.type = "dpsgd"
        self._clip = clip
        self._batch_size = batch_size
        self._sigma = sigma
Z
zhongpu 已提交
2096 2097 2098 2099 2100 2101 2102
        '''
        Note(wangzhongpu):
        This property is only used for debugging, do not need to set it!
        Dpsgd operator use time(NULL) as random seed to generate random number.
        However, during debugging, we need determinated result, so we will set self._seed to a fixed number.
        '''
        self._seed = None
2103 2104 2105 2106 2107

    def _append_optimize_op(self, block, param_and_grad):
        assert isinstance(block, framework.Block)

        # create the dpsgd optimize op
Z
zhongpu 已提交
2108 2109 2110
        if self._seed == None:
            self._seed = 0

2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121
        dpsgd_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
                "LearningRate": self._create_param_lr(param_and_grad)
            },
            outputs={"ParamOut": param_and_grad[0]},
            attrs={
                "clip": self._clip,
                "batch_size": self._batch_size,
Z
zhongpu 已提交
2122 2123
                "sigma": self._sigma,
                "seed": self._seed
2124 2125 2126 2127 2128 2129
            },
            stop_gradient=True)

        return dpsgd_op


2130
class DecayedAdagradOptimizer(Optimizer):
2131
    """
2132 2133 2134
    The Decayed Adagrad optimizer can be seen as an Adagrad algorithm that introduces
    the decay rate to solve the problem of a sharp drop in the learning rate
    during model training when using the AdagradOptimizer.
2135

2136
    The parameter ``param_out`` update rule with gradient ``grad``:
2137 2138 2139 2140 2141 2142 2143

    .. math::

        moment\_out & = decay * moment + (1 - decay) * grad * grad

        param\_out & = param - \\frac{learning\_rate * grad}{\sqrt{moment\_out} + \epsilon}

2144 2145 2146 2147
    Related paper: `Adaptive Subgradient Methods for Online Learning and Stochastic
    Optimization <http://www.jmlr.org/papers/volume12/duchi11a/duchi11a.pdf>`_.

    The original paper does not have an ``epsilon`` attribute. It is added here for numerical
2148 2149 2150
    stability to avoid the division by zero error.

    Args:
2151 2152 2153 2154 2155
        learning_rate (float|Variable): The learning rate used to update ``Parameter``.
            It can be a float value or a ``Variable`` with a float type.
        decay (float, optional): The decay rate. The default value is 0.95.
        epsilon (float, optional): A small float value for numerical stability.
            The default value is 1e-06.
2156 2157 2158
        parameter_list (list, optional):  List of ``Variable`` names to update to minimize ``loss``. \
            This parameter is required in dygraph mode. \
            The default value is None in static mode, at this time all parameters will be updated.
2159 2160 2161 2162 2163 2164 2165 2166
        regularization (WeightDecayRegularizer, optional): A ``Regularizer``, such as
             :ref:`api_fluid_regularizer_L2DecayRegularizer`. The default value is None.
        name (str, optional): Normally there is no need for user to set this property.
            For more information, please refer to :ref:`api_guide_Name`.
            The default value is None.

    **Notes**:
        **Currently, DecayedAdagradOptimizer doesn't support sparse parameter optimization.**
2167 2168 2169 2170

    Examples:
        .. code-block:: python

2171 2172
            import paddle.fluid as fluid

2173 2174 2175 2176
            x = fluid.data( name='x', shape=[None, 10], dtype='float32' )
            trans = fluid.layers.fc( x, 100 )
            cost = fluid.layers.reduce_mean( trans )
            optimizer = fluid.optimizer.DecayedAdagradOptimizer(learning_rate=0.2)
2177
            optimizer.minimize(cost)
2178 2179 2180
    """
    _moment_acc_str = "moment"

X
Xin Pan 已提交
2181 2182 2183 2184
    def __init__(self,
                 learning_rate,
                 decay=0.95,
                 epsilon=1.0e-6,
2185
                 parameter_list=None,
X
Xin Pan 已提交
2186 2187
                 regularization=None,
                 name=None):
2188 2189 2190 2191
        assert learning_rate is not None
        assert decay is not None
        assert epsilon is not None

Q
Qiao Longfei 已提交
2192
        super(DecayedAdagradOptimizer, self).__init__(
X
Xin Pan 已提交
2193
            learning_rate=learning_rate,
2194
            parameter_list=parameter_list,
X
Xin Pan 已提交
2195 2196
            regularization=regularization,
            name=name)
2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223
        self.type = "decayed_adagrad"
        self._decay = decay
        self._epsilon = epsilon

    def _create_accumulators(self, block, parameters):
        assert isinstance(block, framework.Block)

        for p in parameters:
            self._add_accumulator(self._moment_acc_str, p)

    def _append_optimize_op(self, block, param_and_grad):
        assert isinstance(block, framework.Block)

        moment_acc = self._get_accumulator(self._moment_acc_str,
                                           param_and_grad[0])

        # Create the decayed adagrad optimizer op
        decayed_adagrad_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
                "Moment": moment_acc,
                "LearningRate": self._create_param_lr(param_and_grad)
            },
            outputs={"ParamOut": param_and_grad[0],
                     "MomentOut": moment_acc},
2224 2225
            attrs={"epsilon": self._epsilon,
                   "decay": self._decay},
M
minqiyang 已提交
2226
            stop_gradient=True)
2227 2228

        return decayed_adagrad_op
2229 2230


2231
class AdadeltaOptimizer(Optimizer):
2232
    """
Z
Zeng Jinle 已提交
2233
    **Notes: This API does not support sparse parameter optimization.**
Q
qiaolongfei 已提交
2234

Z
Zeng Jinle 已提交
2235
    Adadelta Optimizer. Please refer to this for details:
Z
Zeng Jinle 已提交
2236 2237 2238
    `ADADELTA: AN ADAPTIVE LEARNING RATE METHOD <https://arxiv.org/abs/1212.5701>`_.

    The update is done as follows:
2239

Z
Zeng Jinle 已提交
2240 2241
    .. math::

Z
Zeng Jinle 已提交
2242
        E(g_t^2) &= \\rho * E(g_{t-1}^2) + (1-\\rho) * g^2
2243

Z
Zeng Jinle 已提交
2244
        learning\_rate &= \sqrt{ ( E(dx_{t-1}^2) + \\epsilon ) / ( E(g_t^2) + \\epsilon ) }
Z
Zeng Jinle 已提交
2245

Z
Zeng Jinle 已提交
2246
        E(dx_t^2) &= \\rho * E(dx_{t-1}^2) + (1-\\rho) * (-g*learning\_rate)^2
2247 2248

    Args:
Z
Zeng Jinle 已提交
2249 2250 2251
        learning_rate (float|Variable): global learning rate.
        epsilon (float): a small float number for numeric stability. Default 1.0e-6.
        rho (float): a floating point value indicating the decay rate. Default 0.95.
2252 2253 2254
        parameter_list (list, optional):  List of ``Variable`` names to update to minimize ``loss``. \
            This parameter is required in dygraph mode. \
            The default value is None in static mode, at this time all parameters will be updated.
Z
Zeng Jinle 已提交
2255 2256 2257
        regularization (WeightDecayRegularizer, optional): A Regularizer, such as
                fluid.regularizer.L2DecayRegularizer. Default None, meaning that there is no
                regularization.
2258 2259 2260
        name (str, optional): The default value is None. Normally there is no need for user
                to set this property. For more information, please refer to
                :ref:`api_guide_Name` .
2261 2262 2263 2264

    Examples:
        .. code-block:: python

2265
            import paddle.fluid as fluid
Z
Zeng Jinle 已提交
2266

2267
            image = fluid.data(name='image', shape=[None, 28], dtype='float32')
Z
Zeng Jinle 已提交
2268 2269
            fc = fluid.layers.fc(image, size=10)
            cost = fluid.layers.reduce_mean(fc)
2270 2271
            optimizer = fluid.optimizer.Adadelta(
                learning_rate=0.0003, epsilon=1.0e-6, rho=0.95)
C
chengduo 已提交
2272

Z
Zeng Jinle 已提交
2273 2274 2275 2276
            # optimizer_ops is a list of optimizer operators to update parameters
            # params_grads is a list of (param, param_grad), where param is each
            # parameter and param_grad is the gradient variable of param.
            optimizer_ops, params_grads = optimizer.minimize(cost)
2277
    """
2278

2279 2280 2281
    _avg_squared_grad_acc_str = "_avg_squared_grad"
    _avg_squared_update_acc_str = "_avg_squared_update"

X
Xin Pan 已提交
2282 2283 2284 2285
    def __init__(self,
                 learning_rate,
                 epsilon=1.0e-6,
                 rho=0.95,
2286
                 parameter_list=None,
X
Xin Pan 已提交
2287 2288
                 regularization=None,
                 name=None):
2289 2290 2291 2292 2293 2294
        if learning_rate is None:
            raise ValueError("learning_rate is not set.")
        if epsilon is None:
            raise ValueError("epsilon is not set.")
        if rho is None:
            raise ValueError("rho is not set.")
2295
        super(AdadeltaOptimizer, self).__init__(
X
Xin Pan 已提交
2296
            learning_rate=learning_rate,
2297
            parameter_list=parameter_list,
X
Xin Pan 已提交
2298 2299
            regularization=regularization,
            name=name)
2300 2301 2302 2303 2304
        self.type = "adadelta"
        self._epsilon = epsilon
        self._rho = rho

    def _create_accumulators(self, block, parameters):
2305 2306
        if not isinstance(block, framework.Block):
            raise TypeError("block is not instance of framework.Block.")
2307 2308 2309 2310 2311 2312

        for p in parameters:
            self._add_accumulator(self._avg_squared_grad_acc_str, p)
            self._add_accumulator(self._avg_squared_update_acc_str, p)

    def _append_optimize_op(self, block, param_and_grad):
2313 2314
        if not isinstance(block, framework.Block):
            raise TypeError("block is not instance of framework.Block.")
2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335

        avg_squared_grad_acc = self._get_accumulator(
            self._avg_squared_grad_acc_str, param_and_grad[0])
        avg_squared_update_acc = self._get_accumulator(
            self._avg_squared_update_acc_str, param_and_grad[0])

        # Create the adadelta optimizer op
        adadelta_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
                "AvgSquaredGrad": avg_squared_grad_acc,
                "AvgSquaredUpdate": avg_squared_update_acc
            },
            outputs={
                "ParamOut": param_and_grad[0],
                "AvgSquaredGradOut": avg_squared_grad_acc,
                "AvgSquaredUpdateOut": avg_squared_update_acc
            },
            attrs={"epsilon": self._epsilon,
M
minqiyang 已提交
2336 2337
                   "rho": self._rho},
            stop_gradient=True)
2338 2339 2340 2341

        return adadelta_op


Q
qingqing01 已提交
2342 2343 2344 2345 2346 2347 2348 2349 2350 2351
class RMSPropOptimizer(Optimizer):
    """
    Root Mean Squared Propagation (RMSProp) is an unpublished, adaptive learning
    rate method. The original slides proposed RMSProp: Slide 29 of
    http://www.cs.toronto.edu/~tijmen/csc321/slides/lecture_slides_lec6.pdf .

    The original equation is as follows:

    ..  math::

Q
qiaolongfei 已提交
2352
        r(w, t) & = \\rho r(w, t-1) + (1 - \\rho)(\\nabla Q_{i}(w))^2
Q
qingqing01 已提交
2353 2354 2355 2356

        w & = w - \\frac{\\eta} {\\sqrt{r(w,t) + \\epsilon}} \\nabla Q_{i}(w)

    The first equation calculates moving average of the squared gradient for
Q
qiaolongfei 已提交
2357
    each weight. Then dividing the gradient by :math:`sqrt{v(w,t)}`.
Q
qingqing01 已提交
2358 2359 2360 2361 2362 2363

    In some cases, adding a momentum term :math: `\\beta` is beneficial.
    In our implementation, Nesterov momentum is used:

    ..  math::

Q
qiaolongfei 已提交
2364
        r(w, t) & = \\rho r(w, t-1) + (1 - \\rho)(\\nabla Q_{i}(w))^2
Q
qingqing01 已提交
2365

2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379
        v(w, t) & = \\beta v(w, t-1) + \\frac{\\eta} {\\sqrt{r(w,t) +
            \\epsilon}} \\nabla Q_{i}(w)

        w & = w - v(w, t)

    if centered is True:

    ..  math::

        r(w, t) & = \\rho r(w, t-1) + (1 - \\rho)(\\nabla Q_{i}(w))^2

        g(w, t) & = \\rho g(w, t-1) + (1 - \\rho)\\nabla Q_{i}(w)

        v(w, t) & = \\beta v(w, t-1) + \\frac{\\eta} {\\sqrt{r(w,t) - (g(w, t))^2 +
Q
qingqing01 已提交
2380 2381 2382 2383
            \\epsilon}} \\nabla Q_{i}(w)

        w & = w - v(w, t)

Q
qiaolongfei 已提交
2384
    where, :math:`\\rho` is a hyperparameter and typical values are 0.9, 0.95
Q
qingqing01 已提交
2385 2386 2387 2388 2389
    and so on. :math: `beta` is the momentum term. :math: `\\epsilon` is a
    smoothing term to avoid division by zero, usually set somewhere in range
    from 1e-4 to 1e-8.


2390 2391 2392
    Parameters:
        learning_rate(float): Global learning rate.
        rho(float): rho is :math: `\\rho` in equation, default is 0.95.
Q
qingqing01 已提交
2393
        epsilon(float): :math: `\\epsilon` in equation is smoothing term to
2394
            avoid division by zero, default is 1e-6.
Q
qiaolongfei 已提交
2395
        momentum(float): :math:`\\beta` in equation is the momentum term,
2396
            default is 0.0.
2397 2398 2399 2400
        centered(bool): If True, gradients are normalized by the estimated variance of
            the gradient; if False, by the uncentered second moment. Setting this to
            True may help with training, but is slightly more expensive in terms of
            computation and memory. Defaults to False.
2401 2402 2403
        parameter_list (list, optional):  List of ``Variable`` names to update to minimize ``loss``. \
            This parameter is required in dygraph mode. \
            The default value is None in static mode, at this time all parameters will be updated.
2404 2405 2406 2407
        regularization: A Regularizer, such as :ref:`api_fluid_regularizer_L2DecayRegularizer`. \
            Optional, default is None.
        name (str, optional): This parameter is used by developers to print debugging information. \
            For details, please refer to :ref:`api_guide_Name`. Default is None.
Q
qingqing01 已提交
2408 2409 2410 2411 2412 2413 2414

    Raises:
        ValueError: If learning_rate, rho, epsilon, momentum are None.

    Examples:
          .. code-block:: python

2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439
            import paddle
            import paddle.fluid as fluid
            import numpy as np

            place = fluid.CPUPlace()
            main = fluid.Program()
            with fluid.program_guard(main):
                x = fluid.layers.data(name='x', shape=[13], dtype='float32')
                y = fluid.layers.data(name='y', shape=[1], dtype='float32')
                y_predict = fluid.layers.fc(input=x, size=1, act=None)
                cost = fluid.layers.square_error_cost(input=y_predict, label=y)
                avg_cost = fluid.layers.mean(cost)

                rms_optimizer = fluid.optimizer.RMSProp(learning_rate=0.1)
                rms_optimizer.minimize(avg_cost)

                fetch_list = [avg_cost]
                train_reader = paddle.batch(
                    paddle.dataset.uci_housing.train(), batch_size=1)
                feeder = fluid.DataFeeder(place=place, feed_list=[x, y])
                exe = fluid.Executor(place)
                exe.run(fluid.default_startup_program())
                for data in train_reader():
                    exe.run(main, feed=feeder.feed(data), fetch_list=fetch_list)

Q
qingqing01 已提交
2440 2441 2442 2443
    """

    _momentum_acc_str = "momentum"
    _mean_square_acc_str = "mean_square"
2444
    _mean_grad_acc_str = "mean_grad"
Q
qingqing01 已提交
2445 2446 2447 2448 2449 2450

    def __init__(self,
                 learning_rate,
                 rho=0.95,
                 epsilon=1.0e-6,
                 momentum=0.0,
2451
                 centered=False,
2452
                 parameter_list=None,
X
Xin Pan 已提交
2453 2454
                 regularization=None,
                 name=None):
Q
qingqing01 已提交
2455
        super(RMSPropOptimizer, self).__init__(
X
Xin Pan 已提交
2456
            learning_rate=learning_rate,
2457
            parameter_list=parameter_list,
X
Xin Pan 已提交
2458 2459
            regularization=regularization,
            name=name)
Q
qingqing01 已提交
2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472
        if learning_rate is None:
            raise ValueError("learning_rate is not set.")
        if rho is None:
            raise ValueError("rho is not set.")
        if epsilon is None:
            raise ValueError("epsilon is not set.")
        if momentum is None:
            raise ValueError("momentum is not set.")

        self.type = "rmsprop"
        self._rho = rho
        self._epsilon = epsilon
        self._momentum = momentum
2473
        self._centered = centered
Q
qingqing01 已提交
2474 2475 2476 2477 2478 2479 2480 2481

    def _create_accumulators(self, block, parameters):
        if not isinstance(block, framework.Block):
            raise TypeError("block is not instance of framework.Block.")

        for p in parameters:
            self._add_accumulator(self._momentum_acc_str, p)
            self._add_accumulator(self._mean_square_acc_str, p)
2482
            self._add_accumulator(self._mean_grad_acc_str, p)
Q
qingqing01 已提交
2483 2484 2485 2486 2487 2488 2489 2490 2491

    def _append_optimize_op(self, block, param_and_grad):
        if not isinstance(block, framework.Block):
            raise TypeError("block is not instance of framework.Block.")

        momentum_acc = self._get_accumulator(self._momentum_acc_str,
                                             param_and_grad[0])
        mean_square_acc = self._get_accumulator(self._mean_square_acc_str,
                                                param_and_grad[0])
2492 2493
        mean_grad_acc = self._get_accumulator(self._mean_grad_acc_str,
                                              param_and_grad[0])
Q
qingqing01 已提交
2494 2495 2496 2497 2498 2499 2500
        rmsprop_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
                "Moment": momentum_acc,
                "MeanSquare": mean_square_acc,
2501
                "MeanGrad": mean_grad_acc,
Q
qingqing01 已提交
2502 2503 2504 2505 2506
                "LearningRate": self._create_param_lr(param_and_grad),
            },
            outputs={
                "ParamOut": param_and_grad[0],
                "MomentOut": momentum_acc,
2507 2508
                "MeanSquareOut": mean_square_acc,
                "MeanGradOut": mean_grad_acc
Q
qingqing01 已提交
2509 2510 2511 2512
            },
            attrs={
                "epsilon": self._epsilon,
                "decay": self._rho,
2513 2514
                "momentum": self._momentum,
                "centered": self._centered
M
minqiyang 已提交
2515 2516
            },
            stop_gradient=True)
Q
qingqing01 已提交
2517 2518 2519 2520

        return rmsprop_op


Q
qiaolongfei 已提交
2521 2522 2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560
class FtrlOptimizer(Optimizer):
    """
    FTRL (Follow The Regularized Leader) Optimizer.

    The paper that proposed Follow The Regularized Leader (FTRL):
    (https://www.eecs.tufts.edu/~dsculley/papers/ad-click-prediction.pdf)

    ..  math::

        &new\_accum = squared\_accum + grad^2

        &if (lr\_power == -0.5):

        &\quad  linear\_accum += grad - \\frac{\\sqrt{new\_accum} - \\sqrt{squared\_accum}}{learning\_rate * param}

        &else:

        &\quad   linear\_accum += grad - \\frac{new\_accum^{-lr\_power} - accum^{-lr\_power}}{learning\_rate * param}


        &x = l1 * sign(linear\_accum) - linear\_accum

        &if (lr\_power == -0.5):

        &\quad   y = \\frac{\\sqrt{new\_accum}}{learning\_rate} + (2 * l2)

        &\quad   pre\_shrink = \\frac{x}{y}

        &\quad   param = (abs(linear\_accum) > l1).select(pre\_shrink, 0.0)

        &else:

        &\quad   y = \\frac{new\_accum^{-lr\_power}}{learning\_rate} + (2 * l2)

        &\quad   pre\_shrink = \\frac{x}{y}

        &\quad   param = (abs(linear\_accum) > l1).select(pre\_shrink, 0.0)

        &squared\_accum += grad^2

2561 2562 2563 2564 2565
    Parameters:
        learning_rate (float|Variable): Global learning rate.
        l1 (float): L1 regularization strength, default is 0.0.
        l2 (float): L2 regularization strength, default is 0.0.
        lr_power (float): Learning Rate Power, default is -0.5.
2566 2567 2568
        parameter_list (list, optional):  List of ``Variable`` names to update to minimize ``loss``. \
            This parameter is required in dygraph mode. \
            The default value is None in static mode, at this time all parameters will be updated.
2569 2570 2571 2572
        regularization: A Regularizer, such as :ref:`api_fluid_regularizer_L2DecayRegularizer`. \
            Optional, default is None.
        name (str, optional): This parameter is used by developers to print debugging information. \
            For details, please refer to :ref:`api_guide_Name`. Default is None.
Q
qiaolongfei 已提交
2573 2574 2575 2576 2577 2578 2579

    Raises:
        ValueError: If learning_rate, rho, epsilon, momentum are None.

    Examples:
          .. code-block:: python

2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603
            import paddle
            import paddle.fluid as fluid
            import numpy as np

            place = fluid.CPUPlace()
            main = fluid.Program()
            with fluid.program_guard(main):
                x = fluid.layers.data(name='x', shape=[13], dtype='float32')
                y = fluid.layers.data(name='y', shape=[1], dtype='float32')
                y_predict = fluid.layers.fc(input=x, size=1, act=None)
                cost = fluid.layers.square_error_cost(input=y_predict, label=y)
                avg_cost = fluid.layers.mean(cost)

                ftrl_optimizer = fluid.optimizer.Ftrl(learning_rate=0.1)
                ftrl_optimizer.minimize(avg_cost)

                fetch_list = [avg_cost]
                train_reader = paddle.batch(
                    paddle.dataset.uci_housing.train(), batch_size=1)
                feeder = fluid.DataFeeder(place=place, feed_list=[x, y])
                exe = fluid.Executor(place)
                exe.run(fluid.default_startup_program())
                for data in train_reader():
                    exe.run(main, feed=feeder.feed(data), fetch_list=fetch_list)
C
chengduo 已提交
2604

2605
    NOTE:
C
chengduo 已提交
2606
       Currently, FtrlOptimizer doesn't support sparse parameter optimization.
Q
qiaolongfei 已提交
2607 2608 2609 2610 2611
    """

    _squared_acc_str = "squared"
    _linear_acc_str = "linear"

X
Xin Pan 已提交
2612 2613 2614 2615 2616
    def __init__(self,
                 learning_rate,
                 l1=0.0,
                 l2=0.0,
                 lr_power=-0.5,
2617
                 parameter_list=None,
X
Xin Pan 已提交
2618 2619
                 regularization=None,
                 name=None):
Q
qiaolongfei 已提交
2620
        super(FtrlOptimizer, self).__init__(
X
Xin Pan 已提交
2621
            learning_rate=learning_rate,
2622
            parameter_list=parameter_list,
X
Xin Pan 已提交
2623 2624
            regularization=regularization,
            name=name)
Q
qiaolongfei 已提交
2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664
        if learning_rate is None:
            raise ValueError("learning_rate is not set.")

        self.type = "ftrl"
        self._l1 = l1
        self._l2 = l2
        self._lr_power = lr_power

    def _create_accumulators(self, block, parameters):
        if not isinstance(block, framework.Block):
            raise TypeError("block is not instance of framework.Block.")

        for p in parameters:
            self._add_accumulator(self._squared_acc_str, p)
            self._add_accumulator(self._linear_acc_str, p)

    def _append_optimize_op(self, block, param_and_grad):
        if not isinstance(block, framework.Block):
            raise TypeError("block is not instance of framework.Block.")

        squared_acc = self._get_accumulator(self._squared_acc_str,
                                            param_and_grad[0])
        linear_acc = self._get_accumulator(self._linear_acc_str,
                                           param_and_grad[0])
        ftrl_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
                "SquaredAccumulator": squared_acc,
                "LinearAccumulator": linear_acc,
                "LearningRate": self._create_param_lr(param_and_grad),
            },
            outputs={
                "ParamOut": param_and_grad[0],
                "SquaredAccumOut": squared_acc,
                "LinearAccumOut": linear_acc
            },
            attrs={"l1": self._l1,
                   "l2": self._l1,
M
minqiyang 已提交
2665 2666
                   "lr_power": self._lr_power},
            stop_gradient=True)
Q
qiaolongfei 已提交
2667 2668 2669 2670

        return ftrl_op


Y
Yibing Liu 已提交
2671 2672 2673 2674 2675 2676
class LambOptimizer(AdamOptimizer):
    """
    LAMB (Layer-wise Adaptive Moments optimizer for Batching training) Optimizer.

    LAMB Optimizer is designed to scale up the batch size of training without losing 
    accuracy, which supports adaptive element-wise updating and accurate layer-wise 
Y
Yibing Liu 已提交
2677 2678
    correction. For more information, please refer to `Large Batch Optimization for 
    Deep Learning: Training BERT in 76 minutes <https://arxiv.org/abs/1904.00962>`_ .
Y
Yibing Liu 已提交
2679 2680 2681 2682 2683

    The updating of parameters follows:

    ..  math::

Y
Yibing Liu 已提交
2684
        m_t &= \\beta_1 m_{t - 1}+ (1 - \\beta_1)g_t 
Y
Yibing Liu 已提交
2685

Y
Yibing Liu 已提交
2686
        v_t &= \\beta_2 v_{t - 1}  + (1 - \\beta_2)g_t^2
Y
Yibing Liu 已提交
2687

Y
Yibing Liu 已提交
2688
        r_t &= \\frac{m_t}{\\sqrt{v_t}+\\epsilon}
Y
Yibing Liu 已提交
2689

Y
Yibing Liu 已提交
2690
        w_t &= w_{t-1} -\\eta_t \\frac{\\left \| w_{t-1}\\right \|}{\\left \| r_t + \\lambda w_{t-1}\\right \|} (r_t + \\lambda w_{t-1})
Y
Yibing Liu 已提交
2691 2692 2693 2694 2695 2696


    where :math:`m` is the 1st moment, and :math:`v` the 2nd moment, :math:`\\eta` the 
    learning rate, :math:`\\lambda` the LAMB weight decay rate.

    Args:
Y
Yibing Liu 已提交
2697 2698 2699 2700 2701 2702 2703 2704
        learning_rate (float|Variable, optional): the learning rate used to update parameters. \
            Can be a float value or a Variable with data type float32. Default 0.001.
        lamb_weight_decay (float, optional): The LAMB weight decay rate. Default 0.01.
        beta1 (float, optional): The exponential decay rate for the 1st moment estimates.
            Default 0.9.
        beta2 (float, optional): The exponential decay rate for the 2nd moment estimates.
            Default 0.999.
        epsilon (float, optional): A small float value for numerical stability. Default 1e-6.
2705 2706 2707
        parameter_list (list, optional):  List of ``Variable`` names to update to minimize ``loss``. \
            This parameter is required in dygraph mode. \
            The default value is None in static mode, at this time all parameters will be updated.
Y
Yibing Liu 已提交
2708 2709 2710 2711 2712 2713 2714
        regularization (Regularizer|None): A Regularizer, such as
           fluid.regularizer.L1DecayRegularizer. Default None.
        exclude_from_weight_decay_fn (function|None): Exclude a parameter from weight 
            decay when **exclude_from_weight_decay_fn(parameter)** returns true. 
            Default None.
        name(str|None): For detailed information, please refer to 
            :ref:`api_guide_Name` . Usually name is no need to set and None by default.
Y
Yibing Liu 已提交
2715 2716 2717 2718 2719 2720

    Examples:
        .. code-block:: python
            
            import paddle.fluid as fluid 

Y
Yibing Liu 已提交
2721
            data = fluid.data(name='x', shape=[-1, 5], dtype='float32')
Y
Yibing Liu 已提交
2722 2723 2724
            hidden = fluid.layers.fc(input=data, size=10)
            cost = fluid.layers.mean(hidden)

Y
Yibing Liu 已提交
2725 2726 2727 2728 2729
            def exclude_fn(param):
                return param.name.endswith('.b_0')

            optimizer = fluid.optimizer.Lamb(learning_rate=0.002,
                                             exclude_from_weight_decay_fn=exclude_fn)
Y
Yibing Liu 已提交
2730 2731 2732 2733
            optimizer.minimize(cost)
    """
    _moment1_acc_str = "moment1"
    _moment2_acc_str = "moment2"
Y
Yibing Liu 已提交
2734
    # these two not used in op temporarily
Y
Yibing Liu 已提交
2735 2736 2737 2738 2739 2740 2741 2742 2743
    _beta1_pow_acc_str = "beta1_pow_acc"
    _beta2_pow_acc_str = "beta2_pow_acc"

    def __init__(self,
                 learning_rate=0.001,
                 lamb_weight_decay=0.01,
                 beta1=0.9,
                 beta2=0.999,
                 epsilon=1e-6,
2744
                 parameter_list=None,
Y
Yibing Liu 已提交
2745
                 regularization=None,
Y
Yibing Liu 已提交
2746
                 exclude_from_weight_decay_fn=None,
Y
Yibing Liu 已提交
2747 2748 2749 2750 2751 2752 2753 2754
                 name=None):
        assert learning_rate is not None
        assert lamb_weight_decay is not None
        assert beta1 is not None
        assert beta2 is not None
        assert epsilon is not None
        super(LambOptimizer, self).__init__(
            learning_rate=learning_rate,
2755
            parameter_list=parameter_list,
Y
Yibing Liu 已提交
2756 2757 2758 2759 2760 2761 2762
            regularization=regularization,
            beta1=beta1,
            beta2=beta2,
            epsilon=epsilon,
            name=name)
        self.type = "lamb"
        self._weight_decay = lamb_weight_decay
Y
Yibing Liu 已提交
2763
        self._exclude_from_weight_decay_fn = exclude_from_weight_decay_fn
Y
Yibing Liu 已提交
2764 2765 2766

    def _append_optimize_op(self, block, param_and_grad):
        assert isinstance(block, framework.Block)
2767
        block.program._use_lamb = True
Y
Yibing Liu 已提交
2768 2769 2770 2771 2772 2773 2774 2775 2776 2777

        moment1 = self._get_accumulator(self._moment1_acc_str,
                                        param_and_grad[0])
        moment2 = self._get_accumulator(self._moment2_acc_str,
                                        param_and_grad[0])
        beta1_pow_acc = self._get_accumulator(self._beta1_pow_acc_str,
                                              param_and_grad[0])
        beta2_pow_acc = self._get_accumulator(self._beta2_pow_acc_str,
                                              param_and_grad[0])

Y
Yibing Liu 已提交
2778 2779 2780 2781 2782 2783
        if self._exclude_from_weight_decay_fn is not None \
            and self._exclude_from_weight_decay_fn(param_and_grad[0]):
            weight_decay = 0.0
        else:
            weight_decay = self._weight_decay

Y
Yibing Liu 已提交
2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804
        # create the lamb optimize op
        lamb_op = block.append_op(
            type=self.type,
            inputs={
                "Param": param_and_grad[0],
                "Grad": param_and_grad[1],
                "LearningRate": self._create_param_lr(param_and_grad),
                "Moment1": moment1,
                "Moment2": moment2,
                "Beta1Pow": beta1_pow_acc,
                "Beta2Pow": beta2_pow_acc
            },
            outputs={
                "ParamOut": param_and_grad[0],
                "Moment1Out": moment1,
                "Moment2Out": moment2
            },
            attrs={
                "beta1": self._beta1,
                "beta2": self._beta2,
                "epsilon": self._epsilon,
Y
Yibing Liu 已提交
2805
                "weight_decay": weight_decay
Y
Yibing Liu 已提交
2806 2807 2808 2809 2810 2811
            },
            stop_gradient=True)

        return lamb_op


2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824
# We short the class name, since users will use the optimizer with the package
# name. The sample code:
#
# import paddle.fluid as fluid
#
# sgd = fluid.optimizer.SGD(...)
#
# It is no need to add an `Optimizer` as the class suffix
SGD = SGDOptimizer
Momentum = MomentumOptimizer
Adagrad = AdagradOptimizer
Adam = AdamOptimizer
Adamax = AdamaxOptimizer
2825
Dpsgd = DpsgdOptimizer
2826
DecayedAdagrad = DecayedAdagradOptimizer
2827
Adadelta = AdadeltaOptimizer
Q
qingqing01 已提交
2828
RMSProp = RMSPropOptimizer
Q
qiaolongfei 已提交
2829
Ftrl = FtrlOptimizer
2830
LarsMomentum = LarsMomentumOptimizer
Y
Yibing Liu 已提交
2831
Lamb = LambOptimizer
2832 2833 2834


class ModelAverage(Optimizer):
2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853
    """
    The ModelAverage optimizer accumulates specific continuous historical parameters
    during training. The accumulated historical range can be controlled by the passed
    ``average_window_rate`` argument. The averaged ``Parameter`` are used in the prediction,
    which usually can improve the accuracy of the prediction.

    Accumulate the average of the ``Parameter`` in the sliding window, the result will be saved
    in a temporary variable, can be applied to the current model's ``Parameter`` by calling
    the ``apply()`` method, and the current model ``Parameter`` can be restored by calling
    the ``restore()`` method.

    The window size for calculating the average is determined by ``average_window_rate``,
    ``min_average_window``, ``max_average_window`` and the current ``Parameter`` update times (num_updates).

    When the cumulative times (num_accumulates) is greater than the specific window
    threshold (average_window), the accumulated ``Parameter`` temporary variable is set to 0.0.
    The following example will help to understand the role of these arguments:

    ::
2854

2855 2856 2857 2858 2859 2860 2861 2862 2863
        if num_accumulates >= min_average_window and num_accumulates >= min(max_average_window, num_updates * average_window_rate):
            num_accumulates = 0

    In the above conditional judgment statement, ``num_accumulates`` indicates the current
    accumulated number, which can be abstractly understood as the length of the cumulative window.
    The length of the window must be at least the length set by the ``min_average_window`` argument,
    and cannot exceed the length specified by the ``max_average_window`` argument or
    ``num_updates * average_window_rate``, where ``num_updates`` indicates the current ``Parameter``
    update times, ``average_window_rate`` is a coefficient that calculates the length of the window.
2864 2865

    Args:
2866 2867 2868 2869 2870 2871 2872 2873
        average_window_rate (float): The calculate ratio of the window length relative to ``Parameter`` update times.
        min_average_window (int, optional): the minimum size of average window length. The default value is 10000.
        max_average_window (int, optional): The maximum size of average window length. The default value is 10000.
        regularization (WeightDecayRegularizer, optional): A ``Regularizer``, such as
             :ref:`api_fluid_regularizer_L2DecayRegularizer`. The default value is None.
        name (str, optional): Normally there is no need for user to set this property.
            For more information, please refer to :ref:`api_guide_Name`.
            The default value is None.
2874

2875
    Examples:
Q
qiaolongfei 已提交
2876 2877 2878

      .. code-block:: python

2879 2880 2881 2882 2883 2884
        import paddle.fluid as fluid
        import numpy

        # First create the Executor.
        place = fluid.CPUPlace()  # fluid.CUDAPlace(0)
        exe = fluid.Executor(place)
2885

2886 2887 2888 2889
        train_program = fluid.Program()
        startup_program = fluid.Program()
        with fluid.program_guard(train_program, startup_program):
            # build net
2890
            data = fluid.data(name='X', shape=[None, 1], dtype='float32')
2891 2892 2893 2894 2895 2896 2897 2898
            hidden = fluid.layers.fc(input=data, size=10)
            loss = fluid.layers.mean(hidden)
            optimizer = fluid.optimizer.Momentum(learning_rate=0.2, momentum=0.1)
            optimizer.minimize(loss)

            # build ModelAverage optimizer
            model_average = fluid.optimizer.ModelAverage(0.15,
                                                         min_average_window=10000,
2899
                                                         max_average_window=12500)
2900 2901

            exe.run(startup_program)
2902 2903 2904 2905 2906
            for i in range(12500):
                x = numpy.random.random(size=(10, 1)).astype('float32')
                outs = exe.run(program=train_program,
                               feed={'X': x},
                               fetch_list=[loss.name])
2907 2908

            # apply ModelAverage
2909
            with model_average.apply(exe):
2910 2911 2912 2913
                x = numpy.random.random(size=(10, 1)).astype('float32')
                exe.run(program=train_program,
                        feed={'X': x},
                        fetch_list=[loss.name])
2914 2915 2916
    """

    def __init__(self,
W
wanghaoshuang 已提交
2917
                 average_window_rate,
2918 2919
                 min_average_window=10000,
                 max_average_window=10000,
X
Xin Pan 已提交
2920 2921
                 regularization=None,
                 name=None):
Z
zhongpu 已提交
2922 2923
        if framework.in_dygraph_mode():
            raise Exception("In dygraph, don't support ModelAverage.")
X
Xin Pan 已提交
2924 2925
        super(ModelAverage, self).__init__(
            0.0, regularization=regularization, name=name)
2926 2927 2928
        self.average_window = average_window_rate
        self.min_average_window = min_average_window
        self.max_average_window = max_average_window
2929

2930
        self.params_grads = []
2931 2932
        for param in framework.default_main_program().global_block(
        ).all_parameters():
2933
            if param.do_model_average != False:
2934
                grad = param.block.create_var(
2935 2936
                    name=unique_name.generate_with_ignorable_key(".".join(
                        [param.name, 'tmp'])),
2937 2938
                    dtype=param.dtype,
                    persistable=False,
W
wanghaoshuang 已提交
2939
                    stop_gradient=True)
2940
                self.params_grads.append((param, grad))
2941

2942
        for param, grad in self.params_grads:
2943 2944
            if grad is None:
                continue
X
Xin Pan 已提交
2945 2946
            with param.block.program._optimized_guard(
                [param, grad]), name_scope('move_average'):
2947
                self._append_average_accumulate_op(param)
2948

2949 2950 2951 2952
        self.apply_program = Program()
        block = self.apply_program.global_block()
        with program_guard(main_program=self.apply_program):
            for param_grad in self.params_grads:
2953
                self._add_average_apply_op(block, param_grad)
2954 2955 2956 2957 2958

        self.restore_program = Program()
        block = self.restore_program.global_block()
        with program_guard(main_program=self.restore_program):
            for param_grad in self.params_grads:
2959
                self._add_average_restore_op(block, param_grad)
2960

2961
    def _add_average_apply_op(self, block, param_grad):
L
Luo Tao 已提交
2962 2963 2964 2965 2966 2967
        param = block._clone_variable(param_grad[0])
        grad = block._clone_variable(param_grad[1])
        sum_1 = block._clone_variable(self._get_accumulator('sum_1', param))
        sum_2 = block._clone_variable(self._get_accumulator('sum_2', param))
        sum_3 = block._clone_variable(self._get_accumulator('sum_3', param))
        num_accumulates = block._clone_variable(
2968
            self._get_accumulator('num_accumulates', param))
L
Luo Tao 已提交
2969
        old_num_accumulates = block._clone_variable(
2970
            self._get_accumulator('old_num_accumulates', param))
L
Luo Tao 已提交
2971
        num_updates = block._clone_variable(
2972 2973 2974 2975 2976 2977
            self._get_accumulator('num_updates', param))
        # backup param value to grad
        layers.assign(input=param, output=grad)
        # param = (sum_1 + sum_2 + sum_3) / (num_accumulates + old_num_accumulates)
        tmp = layers.sum(x=[num_accumulates, old_num_accumulates])
        sum = layers.sum(x=[sum_1, sum_2, sum_3])
D
dzhwinter 已提交
2978 2979 2980 2981
        tmp = layers.cast(
            x=tmp, dtype='float32' if self._dtype == None else self._dtype)
        sum = layers.cast(
            x=sum, dtype='float32' if self._dtype == None else self._dtype)
S
sneaxiy 已提交
2982
        ops._elementwise_div(x=sum, y=tmp, out=param)
2983 2984

    def _add_average_restore_op(self, block, param_grad):
L
Luo Tao 已提交
2985 2986
        param = block._clone_variable(param_grad[0])
        grad = block._clone_variable(param_grad[1])
2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023
        layers.assign(input=grad, output=param)

    def _append_average_accumulate_op(self, param):
        self.helper = LayerHelper("average_accumulate")
        sum_1 = self._add_accumulator('sum_1', param)
        sum_2 = self._add_accumulator('sum_2', param)
        sum_3 = self._add_accumulator('sum_3', param)
        num_accumulates = self._add_accumulator(
            'num_accumulates', param, dtype='int64', shape=[1])
        old_num_accumulates = self._add_accumulator(
            'old_num_accumulates', param, dtype='int64', shape=[1])
        num_updates = self._add_accumulator(
            'num_updates', param, dtype='int64', shape=[1])

        self.helper.append_op(
            type='average_accumulates',
            inputs={
                "param": param,
                "in_sum_1": sum_1,
                "in_sum_2": sum_2,
                "in_sum_3": sum_3,
                "in_num_accumulates": num_accumulates,
                "in_old_num_accumulates": old_num_accumulates,
                "in_num_updates": num_updates
            },
            outputs={
                "out_sum_1": sum_1,
                "out_sum_2": sum_2,
                "out_sum_3": sum_3,
                "out_num_accumulates": num_accumulates,
                "out_old_num_accumulates": old_num_accumulates,
                "out_num_updates": num_updates,
            },
            attrs={
                "average_window": self.average_window,
                "min_average_window": self.min_average_window,
                "max_average_window": self.max_average_window,
M
minqiyang 已提交
3024 3025
            },
            stop_gradient=True)
3026

S
rename  
sneaxiy 已提交
3027
    @signature_safe_contextmanager
3028
    def apply(self, executor, need_restore=True):
3029 3030
        """
        Apply the average of the cumulative ``Parameter`` to the parameters of the current model.
3031 3032

        Args:
3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076
            executor(fluid.Executor): The current network executor.
            need_restore(bool): Restore flag variable, if set to True, the network will restore
                the parameters of the network to the default value, if set to False,
                it will not be restored. The default value is True.

        Examples:

          .. code-block:: python

            import paddle.fluid as fluid
            import numpy

            # First create the Executor.
            place = fluid.CPUPlace()  # fluid.CUDAPlace(0)
            exe = fluid.Executor(place)

            train_program = fluid.Program()
            startup_program = fluid.Program()
            with fluid.program_guard(train_program, startup_program):
                # build net
                data = fluid.data(name='X', shape=[None, 1], dtype='float32')
                hidden = fluid.layers.fc(input=data, size=10)
                loss = fluid.layers.mean(hidden)
                optimizer = fluid.optimizer.Momentum(learning_rate=0.2, momentum=0.1)
                optimizer.minimize(loss)

                # build ModelAverage optimizer
                model_average = fluid.optimizer.ModelAverage(0.15,
                                                            min_average_window=10000,
                                                            max_average_window=12500)

                exe.run(startup_program)
                for i in range(12500):
                    x = numpy.random.random(size=(10, 1)).astype('float32')
                    outs = exe.run(program=train_program,
                                feed={'X': x},
                                fetch_list=[loss.name])

                # apply ModelAverage
                with model_average.apply(exe):
                    x = numpy.random.random(size=(10, 1)).astype('float32')
                    exe.run(program=train_program,
                            feed={'X': x},
                            fetch_list=[loss.name])
3077
        """
3078 3079 3080 3081 3082 3083
        executor.run(self.apply_program)
        try:
            yield
        finally:
            if need_restore:
                self.restore(executor)
3084 3085

    def restore(self, executor):
3086 3087
        """
        Restore ``Parameter`` values of current model.
3088 3089
        
        Args:
3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133
            executor(fluid.Executor): The current network executor.

        Examples:

          .. code-block:: python

            import paddle.fluid as fluid
            import numpy

            # First create the Executor.
            place = fluid.CPUPlace()  # fluid.CUDAPlace(0)
            exe = fluid.Executor(place)

            train_program = fluid.Program()
            startup_program = fluid.Program()
            with fluid.program_guard(train_program, startup_program):
                # build net
                data = fluid.data(name='X', shape=[None, 1], dtype='float32')
                hidden = fluid.layers.fc(input=data, size=10)
                loss = fluid.layers.mean(hidden)
                optimizer = fluid.optimizer.Momentum(learning_rate=0.2, momentum=0.1)
                optimizer.minimize(loss)

                # build ModelAverage optimizer
                model_average = fluid.optimizer.ModelAverage(0.15,
                                                            min_average_window=10000,
                                                            max_average_window=12500)

                exe.run(startup_program)
                for i in range(12500):
                    x = numpy.random.random(size=(10, 1)).astype('float32')
                    outs = exe.run(program=train_program,
                                feed={'X': x},
                                fetch_list=[loss.name])

                # apply ModelAverage
                with model_average.apply(exe, False):
                    x = numpy.random.random(size=(10, 1)).astype('float32')
                    exe.run(program=train_program,
                            feed={'X': x},
                            fetch_list=[loss.name])

                # restore Parameters
                model_average.restore(exe)
3134
        """
3135
        executor.run(self.restore_program)
3136 3137 3138 3139 3140 3141 3142 3143 3144 3145


class ExponentialMovingAverage(object):
    """
    Compute the moving average of parameters with exponential decay.
    Given a parameter :math:`\\theta`, its exponential moving average (EMA)
    will be

    ..  math::

3146
        \\text{EMA}_0 & = 0
3147

3148 3149
	\\text{EMA}_t & = \\text{decay} * \\text{EMA}_{t-1} + (1 - \\text{decay}) * \\theta_t

Y
Yibing Liu 已提交
3150 3151 3152 3153
    The average results calculated by **update()** method will be saved in 
    temporary variables which are created and maintained by the object, and can 
    be applied to parameters of current model by calling **apply()** method. And 
    the **restore()** method is used to restore the parameters.
3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165 3166 3167 3168 3169 3170 3171 3172 3173 3174

    **Bias correction**. All EMAs are initialized to :math:`0` and hence they will be 
    zero biased, which can be corrected by divided by a factor 
    :math:`(1 - \\text{decay}^t)` , i.e., the actual EMAs applied to parameters 
    when calling **apply()** method would be 

    ..  math::
    
        \\widehat{\\text{EMA}}_t = \\frac{\\text{EMA}_t}{1 - \\text{decay}^t}

    **Decay rate scheduling**. A large decay rate very close to 1 would result 
    in that the averages move very slowly. And a better strategy is to set a 
    relative smaller decay rate in the very beginning. The argument **thres_steps**
    allows users to pass a Variable to schedule the decay rate, in this case, 
    the actual decay rate becomes
     
    ..  math::
    
        \\min(\\text{decay}, \\frac{1 + \\text{thres_steps}}{10 + \\text{thres_steps}})

    Usually **thres_steps** can be the global training steps.
3175 3176 3177


    Args:
Y
Yibing Liu 已提交
3178 3179 3180 3181 3182 3183 3184
	decay (float, optional): The exponential decay rate, usually close to 1, such as 
            0.999, 0.9999, ... . Default 0.999.
        thres_steps (Variable|None): If not `None`, schedule the decay rate. 
            Default None.
        name (str|None): For detailed information, please refer to 
            :ref:`api_guide_Name`. Usually name is no need to set and None by 
            default.
3185 3186 3187 3188 3189


    Examples:

	.. code-block:: python
3190 3191 3192 3193 3194

	    import numpy
	    import paddle
	    import paddle.fluid as fluid

Y
Yibing Liu 已提交
3195
	    data = fluid.data(name='x', shape=[-1, 5], dtype='float32')
3196 3197 3198 3199 3200 3201 3202 3203
	    hidden = fluid.layers.fc(input=data, size=10)
	    cost = fluid.layers.mean(hidden)

	    test_program = fluid.default_main_program().clone(for_test=True)

	    optimizer = fluid.optimizer.Adam(learning_rate=0.001)
	    optimizer.minimize(cost)

3204
	    global_steps = fluid.layers.autoincreased_step_counter()
3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 3232 3233
	    ema = fluid.optimizer.ExponentialMovingAverage(0.999, thres_steps=global_steps)
	    ema.update()

	    place = fluid.CPUPlace()
	    exe = fluid.Executor(place)
	    exe.run(fluid.default_startup_program())

	    for pass_id in range(3):
		for batch_id in range(6):
		    data = numpy.random.random(size=(10, 5)).astype('float32')
		    exe.run(program=fluid.default_main_program(),
			feed={'x': data}, 
			fetch_list=[cost.name])

		# usage 1
		with ema.apply(exe):
		    data = numpy.random.random(size=(10, 5)).astype('float32')
		    exe.run(program=test_program,
			    feed={'x': data}, 
			    fetch_list=[hidden.name])
			    

		 # usage 2
		with ema.apply(exe, need_restore=False):
		    data = numpy.random.random(size=(10, 5)).astype('float32')
		    exe.run(program=test_program,
			    feed={'x': data}, 
			    fetch_list=[hidden.name])
		ema.restore(exe)
3234 3235
    """

3236
    def __init__(self, decay=0.999, thres_steps=None, name=None):
Z
zhongpu 已提交
3237 3238 3239
        if framework.in_dygraph_mode():
            raise Exception(
                "In dygraph, don't support ExponentialMovingAverage.")
3240
        self._decay = decay
3241
        self._thres_steps = thres_steps
3242
        self._name = name if name is not None else ''
3243 3244
        self._decay_var = self._get_ema_decay()

3245
        self._step_counter_name = "@EMA_STEP_COUNTER@"
Y
Yibing Liu 已提交
3246
        self._params_tmps = []
3247
        for param in default_main_program().global_block().all_parameters():
3248 3249 3250 3251 3252 3253 3254
            if param.do_model_average != False:
                tmp = param.block.create_var(
                    name=unique_name.generate(".".join(
                        [self._name + param.name, 'ema_tmp'])),
                    dtype=param.dtype,
                    persistable=False,
                    stop_gradient=True)
Y
Yibing Liu 已提交
3255
                self._params_tmps.append((param, tmp))
3256

Y
Yibing Liu 已提交
3257 3258
        self._ema_vars = {}
        for param, tmp in self._params_tmps:
3259 3260
            with param.block.program._optimized_guard(
                [param, tmp]), name_scope('moving_average'):
Y
Yibing Liu 已提交
3261
                self._ema_vars[param.name] = self._create_ema_vars(param)
3262 3263 3264 3265

        self.apply_program = Program()
        block = self.apply_program.global_block()
        with program_guard(main_program=self.apply_program):
3266
            decay_pow, global_step = self._get_decay_pow(block)
Y
Yibing Liu 已提交
3267
            for param, tmp in self._params_tmps:
3268 3269
                param = block._clone_variable(param)
                tmp = block._clone_variable(tmp)
Y
Yibing Liu 已提交
3270
                ema = block._clone_variable(self._ema_vars[param.name])
3271
                layers.assign(input=param, output=tmp)
3272
                # bias correction
3273 3274 3275
                with layers.control_flow.Switch() as switch:
                    with switch.case(global_step > 0):
                        layers.assign(output=ema, input=ema / (1.0 - decay_pow))
3276 3277 3278 3279 3280
                layers.assign(input=ema, output=param)

        self.restore_program = Program()
        block = self.restore_program.global_block()
        with program_guard(main_program=self.restore_program):
Y
Yibing Liu 已提交
3281
            for param, tmp in self._params_tmps:
3282 3283 3284 3285
                tmp = block._clone_variable(tmp)
                param = block._clone_variable(param)
                layers.assign(input=tmp, output=param)

3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307
    def _get_ema_decay(self):
        with default_main_program()._lr_schedule_guard():
            decay_var = layers.tensor.create_global_var(
                shape=[1],
                value=self._decay,
                dtype='float32',
                persistable=True,
                name="scheduled_ema_decay_rate")

            if self._thres_steps is not None:
                decay_t = (self._thres_steps + 1.0) / (self._thres_steps + 10.0)
                with layers.control_flow.Switch() as switch:
                    with switch.case(decay_t < self._decay):
                        layers.tensor.assign(decay_t, decay_var)
                    with switch.default():
                        layers.tensor.assign(
                            np.array(
                                [self._decay], dtype=np.float32),
                            decay_var)
        return decay_var

    def _get_decay_pow(self, block):
3308 3309 3310 3311 3312 3313 3314
        global_step = layers.create_global_var(
            name=self._step_counter_name,
            shape=[1],
            value=0,
            dtype='int64',
            persistable=True)
        global_step = layers.cast(global_step, "float32")
3315
        decay_var = block._clone_variable(self._decay_var)
3316 3317
        decay_pow_acc = layers.elementwise_pow(decay_var, global_step)
        return decay_pow_acc, global_step
3318

Y
Yibing Liu 已提交
3319
    def _create_ema_vars(self, param):
3320 3321 3322 3323 3324 3325 3326 3327 3328
        param_ema = layers.create_global_var(
            name=unique_name.generate(self._name + param.name + '_ema'),
            shape=param.shape,
            value=0.0,
            dtype=param.dtype,
            persistable=True)

        return param_ema

Y
Yibing Liu 已提交
3329 3330 3331 3332 3333
    def update(self):
        """ 
        Update Exponential Moving Average. Should only call this method in 
        train program.
        """
3334 3335
        global_step = layers.autoincreased_step_counter(
            counter_name=self._step_counter_name)
3336
        param_master_emas = []
Y
Yibing Liu 已提交
3337 3338 3339 3340
        for param, tmp in self._params_tmps:
            with param.block.program._optimized_guard(
                [param, tmp]), name_scope('moving_average'):
                param_ema = self._ema_vars[param.name]
3341
                if param.name + '.master' in self._ema_vars:
3342 3343 3344 3345 3346 3347 3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358
                    master_ema = self._ema_vars[param.name + '.master']
                    param_master_emas.append([param_ema, master_ema])
                else:
                    ema_t = param_ema * self._decay_var + param * (
                        1 - self._decay_var)
                    layers.assign(input=ema_t, output=param_ema)

        # for fp16 params
        for param_ema, master_ema in param_master_emas:
            default_main_program().global_block().append_op(
                type="cast",
                inputs={"X": master_ema},
                outputs={"Out": param_ema},
                attrs={
                    "in_dtype": master_ema.dtype,
                    "out_dtype": param_ema.dtype
                })
Y
Yibing Liu 已提交
3359

3360 3361 3362 3363 3364 3365 3366
    @signature_safe_contextmanager
    def apply(self, executor, need_restore=True):
        """
        Apply moving average to parameters for evaluation.
        
        Args:
            executor (Executor): The Executor to execute applying.
Y
Yibing Liu 已提交
3367 3368
            need_restore (bool, optional): Whether to restore parameters after 
                applying. Default True.
3369 3370 3371 3372 3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383
        """
        executor.run(self.apply_program)
        try:
            yield
        finally:
            if need_restore:
                self.restore(executor)

    def restore(self, executor):
        """Restore parameters.
        
        Args:
            executor (Executor): The Executor to execute restoring.
        """
        executor.run(self.restore_program)
H
hutuxian 已提交
3384 3385 3386


class PipelineOptimizer(object):
3387 3388
    """
    Pipeline Optimizer
H
hutuxian 已提交
3389

T
tianshuo78520a 已提交
3390
    Train with pipeline mode. The program will be split by cut_list. 
H
hutuxian 已提交
3391 3392

    If the len of cut_list is k, then the whole program (including \
T
tianshuo78520a 已提交
3393
    backward part) will be split to 2*k-1 sections. 
H
hutuxian 已提交
3394 3395 3396 3397
    
    So the length of place_list and concurrency_list must be also 2*k-1.

    Note: Though the asynchronous mode is applied in pipeline training to speed up, \
3398
    the final performance depends on the training progress of each pipeline heavily.
H
hutuxian 已提交
3399 3400 3401

    And we will try the synchronous mode in the future.

3402
    Args:
H
hutuxian 已提交
3403 3404 3405 3406
        optimizer (Optimizer): The based optimizer, such as SGD.
        cut_list (list of Variable list): The cut variable of the main_program.
        place_list (list of Place): The place where the section will run on.
        concurrency_list (list of int): The concurrency degree.
3407 3408
        queue_size (int): Each section will consume scopes from its in-scope queue 
                        and produce scopes to out-scope queue. And this parameter 
H
hutuxian 已提交
3409 3410 3411 3412
                        specify the scope queue size. [Optional. Default: 30].
        sync_steps (int): The synchronization steps between different cards. [Optional. Default: 1].
        start_cpu_core_id (int): specify the first cpu core id. [Optional. Default:0].

3413 3414
    Examples:
        .. code-block:: python
H
hutuxian 已提交
3415

3416
            import paddle.fluid as fluid
H
hutuxian 已提交
3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449 3450
            import paddle.fluid.layers as layers

            x = fluid.layers.data(name='x', shape=[1], dtype='int64', lod_level=0)
            y = fluid.layers.data(name='y', shape=[1], dtype='int64', lod_level=0)
            emb_x = layers.embedding(input=x, param_attr=fluid.ParamAttr(name="embx"), size=[10,2], is_sparse=False)
            emb_y = layers.embedding(input=y, param_attr=fluid.ParamAttr(name="emby",learning_rate=0.9), size=[10,2], is_sparse=False)
            concat = layers.concat([emb_x, emb_y], axis=1)
            fc = layers.fc(input=concat, name="fc", size=1, num_flatten_dims=1, bias_attr=False)
            loss = layers.reduce_mean(fc)
            optimizer = fluid.optimizer.SGD(learning_rate=0.5)
            optimizer = fluid.optimizer.PipelineOptimizer(optimizer,
                    cut_list=[[emb_x, emb_y], [loss]],
                    place_list=[fluid.CPUPlace(), fluid.CUDAPlace(0), fluid.CPUPlace()],
                    concurrency_list=[1, 1, 4],
                    queue_size=2,
                    sync_steps=1,
                    )
            optimizer.minimize(loss)
            place = fluid.CPUPlace()
            exe = fluid.Executor(place)
            exe.run(fluid.default_startup_program())
            filelist = [] # you should set your own filelist, e.g. filelist = ["dataA.txt"]
            dataset = fluid.DatasetFactory().create_dataset("FileInstantDataset")
            dataset.set_use_var([x,y])
            dataset.set_batch_size(batch_size)
            dataset.set_filelist(filelist)
            exe.train_from_dataset(
                        fluid.default_main_program(),
                        dataset,
                        thread=2,
                        debug=False,
                        fetch_list=[],
                        fetch_info=[],
                        print_period=1)
3451 3452
    """

H
hutuxian 已提交
3453 3454 3455 3456 3457 3458 3459 3460
    def __init__(self,
                 optimizer,
                 cut_list=None,
                 place_list=None,
                 concurrency_list=None,
                 queue_size=30,
                 sync_steps=1,
                 start_cpu_core_id=0):
Z
zhongpu 已提交
3461 3462
        if framework.in_dygraph_mode():
            raise Exception("In dygraph, don't support PipelineOptimizer.")
H
hutuxian 已提交
3463 3464 3465 3466 3467 3468 3469 3470 3471
        # TODO: check properties
        self._optimizer = optimizer
        self._cut_list = cut_list
        self._place_list = place_list
        self._concurrency_list = concurrency_list
        self._queue_size = queue_size
        self._sync_steps = sync_steps
        self._start_cpu_core_id = start_cpu_core_id

H
hutuxian 已提交
3472
    def _create_vars(self, block, main_program):
H
hutuxian 已提交
3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483
        used_var_set = set()
        for op_idx in range(block.desc.op_size()):
            op_desc = block.desc.op(op_idx)
            vars = op_desc.input_arg_names() + op_desc.output_arg_names()
            for var in vars:
                if var in used_var_set:
                    continue
                used_var_set.add(var)
                source_var = main_program.block(0).var(str(var))
                block._clone_variable(source_var, False)

H
hutuxian 已提交
3484
    def _extract_section_opt_ops(self, ops, cut_point_name):
H
hutuxian 已提交
3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499
        """
        Extract opt ops in the given section
        """
        output_names = set(cut_point_name)
        relevant_op_flags = [True] * len(ops)
        for i, op in reversed(list(enumerate(ops))):
            if _some_in_set_(op.desc.output_arg_names(), output_names):
                for name in op.desc.input_arg_names():
                    output_names.add(name)
            else:
                relevant_op_flags[i] = False

        op_path = [ops[i] for i in range(len(ops)) if relevant_op_flags[i]]
        return op_path

H
hutuxian 已提交
3500
    def _find_input_output(self, ops, name, is_forward=True):
H
hutuxian 已提交
3501 3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514
        """
        Find the inputs or outputs of a section
        """
        all_set = set()
        part_set = set()
        for op in ops:
            if is_forward:
                part_set.update(op.desc.output_arg_names())
            else:
                part_set.update(op.desc.input_arg_names())
            all_set.update(op.desc.output_arg_names())
            all_set.update(op.desc.input_arg_names())
        return all_set - part_set

H
hutuxian 已提交
3515
    def _find_persistable_vars(self, ops, whole_parameters):
H
hutuxian 已提交
3516 3517 3518 3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542
        """
        find the persistable input vars in current section
        """
        res = set()
        for op in ops:
            vars = op.desc.input_arg_names()
            for var in vars:
                if var in whole_parameters:
                    res.add(var)
        return res

    def _is_opt_role_op(self, op):
        op_maker = core.op_proto_and_checker_maker
        optimize_role = core.op_proto_and_checker_maker.OpRole.Optimize
        if op_maker.kOpRoleAttrName() in op.attr_names and \
                int(op.all_attrs()[op_maker.kOpRoleAttrName()]) & int(optimize_role) != 0:
            return True
        return False

    def _is_lr_role_op(self, op):
        op_maker = core.op_proto_and_checker_maker
        optimize_role = core.op_proto_and_checker_maker.OpRole.LRSched
        if op_maker.kOpRoleAttrName() in op.attr_names and \
                int(op.all_attrs()[op_maker.kOpRoleAttrName()]) == int(optimize_role):
            return True
        return False

H
hutuxian 已提交
3543
    def _extract_section_ops(self, ops, cut_point_name):
H
hutuxian 已提交
3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562
        """
        Extract ops in the given section 
        """
        output_names = set(cut_point_name)
        relevant_op_flags = [True] * len(ops)
        for i, op in reversed(list(enumerate(ops))):
            if not self._is_opt_role_op(op) and _some_in_set_(
                    op.desc.output_arg_names(), output_names):
                for name in op.desc.input_arg_names():
                    output_names.add(name)
            elif op.desc.type() == "print" and op.desc.input_arg_names()[
                    0] in output_names:
                continue
            else:
                relevant_op_flags[i] = False

        op_path = [ops[i] for i in range(len(ops)) if relevant_op_flags[i]]
        return op_path

H
hutuxian 已提交
3563 3564
    def _find_section_opt(self, ops, params):
        res = self._extract_section_opt_ops(ops, params)
H
hutuxian 已提交
3565 3566
        return res

H
hutuxian 已提交
3567
    def _split_program(self, main_program, cut_list):
H
hutuxian 已提交
3568 3569 3570 3571 3572 3573 3574 3575 3576 3577 3578 3579 3580 3581 3582 3583 3584 3585 3586 3587
        programs = []
        block = main_program.block(0)
        whole_parameters = [e.name for e in block.all_parameters()]
        cut_var_names = []
        cut_len = len(cut_list)
        sec_params = []
        for i, cut_vars in enumerate(cut_list[:-1]):
            cut_var_names.append([cut_var.name for cut_var in cut_vars])
        for i, cut_vars in reversed(list(enumerate(cut_list[:-1]))):
            cut_var_names.append(
                [_append_grad_suffix_(cut_var.name) for cut_var in cut_vars])
            if i == 0:
                cut_var_names[-1] += [var.name for var in cut_list[-1]]
        ops = block.ops[:]
        for i, cut_vars in enumerate(cut_var_names):
            program = {
                "program": Program(),
                "input_set": set(),
                "output_set": set()
            }
H
hutuxian 已提交
3588
            cur_ops = self._extract_section_ops(ops, cut_vars)
H
hutuxian 已提交
3589 3590 3591 3592 3593 3594
            if i == 0:
                for op in ops:
                    if self._is_lr_role_op(op):
                        cur_ops.append(op)
            #prevent inplace in/out
            program["input_set"].update(
H
hutuxian 已提交
3595
                self._find_input_output(
H
hutuxian 已提交
3596 3597 3598 3599 3600 3601
                    cur_ops, [], is_forward=True))
            for e in cur_ops:
                ops.remove(e)

            if i < cut_len:
                sec_params.append(
H
hutuxian 已提交
3602
                    self._find_persistable_vars(cur_ops, whole_parameters))
H
hutuxian 已提交
3603
            if i >= cut_len - 1:
H
hutuxian 已提交
3604 3605
                opt_ops = self._find_section_opt(
                    ops, sec_params[2 * cut_len - 2 - i])
H
hutuxian 已提交
3606 3607 3608 3609 3610 3611 3612 3613 3614 3615

                for e in opt_ops:
                    ops.remove(e)
                cur_ops += opt_ops

            op_descs = [op.desc for op in cur_ops]
            for op_desc in op_descs:
                ap_op = program["program"].block(0).desc.append_op()
                ap_op.copy_from(op_desc)
            program["input_set"].update(
H
hutuxian 已提交
3616
                self._find_input_output(
H
hutuxian 已提交
3617 3618 3619
                    cur_ops, cut_vars, is_forward=True))
            program["input_set"].update(sec_params[min(i, 2 * cut_len - 2 - i)])
            program["output_set"].update(
H
hutuxian 已提交
3620
                self._find_input_output(
H
hutuxian 已提交
3621 3622 3623 3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634
                    cur_ops, cut_vars, is_forward=False))
            programs.append(program)
        program = {
            "program": Program(),
            "input_set": set(),
            "output_set": set()
        }
        op_descs = [op.desc for op in ops]
        for op_desc in op_descs:
            ap_op = program["program"].block(0).desc.append_op()
            ap_op.copy_from(op_desc)
        program["input_set"].update(
            [cut_var.name + "@GRAD" for cut_var in cut_list[0]])
        program["input_set"].update(
H
hutuxian 已提交
3635
            self._find_input_output(
H
hutuxian 已提交
3636 3637 3638 3639 3640 3641 3642 3643 3644 3645 3646 3647 3648 3649 3650 3651 3652 3653 3654 3655
                ops, [], is_forward=True))
        program["input_set"].update(sec_params[0])
        programs.append(program)
        inputs = set()
        for program in reversed(list(programs)):
            output_list = list(program["output_set"])
            for output in output_list:
                if output not in inputs:
                    program["output_set"].remove(output)
            inputs.update(program["input_set"])
        return programs

    def minimize(self,
                 loss,
                 startup_program=None,
                 parameter_list=None,
                 no_grad_set=None):
        self._optimizer.minimize(loss, startup_program, parameter_list,
                                 no_grad_set)
        program = loss.block.program
H
hutuxian 已提交
3656 3657 3658 3659 3660 3661 3662 3663
        if len(self._cut_list) == 0:
            program_list = []
            ptmp = {"program": program, "input_set": set(), "output_set": set()}
            program_list.append(ptmp)
        else:
            program_list = self._split_program(program, self._cut_list)
            for p in program_list:
                self._create_vars(p["program"].block(0), program)
H
hutuxian 已提交
3664 3665 3666 3667 3668 3669 3670 3671 3672 3673 3674 3675 3676 3677 3678 3679 3680 3681 3682 3683
        whole_parameters = [e.name for e in program.block(0).all_parameters()]
        param_need_sync = []
        for i, section_p in enumerate(program_list):
            if not isinstance(self._place_list[i], core.CUDAPlace):
                continue
            section_var = [e for e in section_p["program"].block(0).vars]
            for p in section_var:
                if p in whole_parameters:
                    param_need_sync.append(p)
        program._pipeline_opt = {
            "trainer": "PipelineTrainer",
            "device_worker": "Section",
            "section_program_list": program_list,
            "place_list": self._place_list,
            "concurrency_list": self._concurrency_list,
            "queue_size": self._queue_size,
            "start_cpu_core_id": self._start_cpu_core_id,
            "sync_steps": self._sync_steps,
            "param_need_sync": param_need_sync
        }
M
mapingshuo 已提交
3684 3685


M
mapingshuo 已提交
3686 3687 3688 3689 3690 3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714 3715 3716 3717 3718 3719 3720 3721 3722 3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740 3741 3742 3743 3744 3745 3746 3747
class RecomputeOptimizer(Optimizer):
    """
    Recompute Optimizer Wrapper

    Normally, a training step contains three sub-steps: first, run forward
    Operators to calculate the loss; second, run backward Operators to 
    calculate gradient of the parameters; third, apply optimization method
    to update the value of the parameters.

    In the forward computation process, all variables that are needed by 
    backward computation process will be kept in memory, which occupy a great
    amount of memory when the network becomes very deep.

    Recompute split the network to k segments. In each segment, It will 
    recompute the forward Operators, before running backward operators. It is
    very helpful for saving memory.
 
    The Variables that separate a network to segments are called as checkpoints,
    and users should set it manually. The usage is very simple:

    Args:
        optimizer (Optimizer): The optimizer that is applied to parameters.

    Examples:
        .. code-block:: python

            import paddle.fluid as fluid
            import numpy as np
            def gen_data():
                return {"x": np.random.random(size=(32, 32)).astype('float32'),
                "y": np.random.randint(2, size=(32, 1)).astype('int64')}
            def mlp(input_x, input_y, hid_dim=128, label_dim=2):
                print(input_x)
                fc_1 = fluid.layers.fc(input=input_x, size=hid_dim)
                prediction = fluid.layers.fc(input=[fc_1], size=label_dim, act='softmax')
                cost = fluid.layers.cross_entropy(input=prediction, label=input_y)
                sum_cost = fluid.layers.reduce_mean(cost)
                return sum_cost, fc_1, prediction
            input_x = fluid.layers.data(name="x", shape=[32], dtype='float32')
            input_y = fluid.layers.data(name="y", shape=[1], dtype='int64')
            cost, fc_1, pred = mlp(input_x, input_y)

            sgd = fluid.optimizer.Adam(learning_rate=0.01)
            sgd = fluid.optimizer.RecomputeOptimizer(sgd)
            sgd._set_checkpoints([fc_1, pred])
            sgd.minimize(cost)

            print("Finished optimize")
            place = fluid.CPUPlace()
            exe = fluid.Executor(place)
            exe.run(fluid.default_startup_program())
            step = 10

            for i in range(step):
                cost_val = exe.run(feed=gen_data(),
                       program=fluid.default_main_program(),
                       fetch_list=[cost.name])
                print("step=%d cost=%f" % (i, cost_val[0]))

    """

    def __init__(self, optimizer):
Z
zhongpu 已提交
3748 3749
        if framework.in_dygraph_mode():
            raise Exception("In dygraph, don't support RecomputeOptimizer.")
M
mapingshuo 已提交
3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760 3761 3762 3763 3764 3765 3766 3767 3768 3769 3770 3771 3772 3773 3774 3775 3776 3777 3778 3779 3780 3781 3782 3783 3784 3785 3786 3787 3788 3789 3790 3791 3792 3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836 3837 3838 3839 3840 3841 3842 3843 3844 3845 3846 3847 3848 3849 3850 3851 3852 3853 3854
        self._optimizer = optimizer
        self._checkpoints = None

    def _set_checkpoints(self, checkpoints):
        self._checkpoints = checkpoints

    def load(self, stat_dict):
        """
        load function is not supported by Recompute Optimizer for now.
        :return: None

        Args:
            stat_dict: the dict load by load_persistable method

        Examples:
            .. code-block:: python

                import paddle.fluid as fluid
                import paddle.compat as cpt
                
                def mlp(input_x, input_y, hid_dim=128, label_dim=2):
                    fc_1 = fluid.layers.fc(input=input_x, size=hid_dim)
                    prediction = fluid.layers.fc(input=[fc_1], size=label_dim, act='softmax')
                    cost = fluid.layers.cross_entropy(input=prediction, label=input_y)
                    sum_cost = fluid.layers.reduce_mean(cost)
                    return sum_cost, fc_1, prediction
                
                input_x = fluid.layers.data(name="x", shape=[32], dtype='float32')
                input_y = fluid.layers.data(name="y", shape=[1], dtype='int64')
                cost, fc_1, pred = mlp(input_x, input_y)
                print("Finished FF")
                
                sgd = fluid.optimizer.Adam(learning_rate=0.01)
                sgd = fluid.optimizer.RecomputeOptimizer(sgd)
                sgd._set_checkpoints([fc_1, pred])
                try:
                    stat_dict = {}
                    sgd.load(stat_dict)
                except NotImplementedError as e:
                    print(cpt.get_exception_message(e))
        """
        raise NotImplementedError(
            "load function is not supported by Recompute Optimizer for now")

    def apply_gradients(self, params_grads):
        """
        call apply_gradients function of self._optimizer.

        Args:
            params_grads (list): list of (param, grad) pair to do optimization.

        Returns:
            list: A list of operators appended to the current program.

        Examples:
            .. code-block:: python

                import paddle.fluid as fluid
                import paddle.fluid.framework as framework

                def mlp(input_x, input_y, hid_dim=128, label_dim=2):
                    fc_1 = fluid.layers.fc(input=input_x, size=hid_dim)
                    prediction = fluid.layers.fc(input=[fc_1], size=label_dim, act='softmax')
                    cost = fluid.layers.cross_entropy(input=prediction, label=input_y)
                    sum_cost = fluid.layers.reduce_mean(cost)
                    return sum_cost, fc_1, prediction


                input_x = fluid.layers.data(name="x", shape=[32], dtype='float32')
                input_y = fluid.layers.data(name="y", shape=[1], dtype='int64')
                cost, fc_1, pred = mlp(input_x, input_y)
                print("Finished FF")

                sgd = fluid.optimizer.Adam(learning_rate=0.01)
                sgd = fluid.optimizer.RecomputeOptimizer(sgd)
                params_grads = sgd.backward(
                    cost,
                    startup_program=None,
                    parameter_list=None,
                    no_grad_set=None,
                    checkpoints=[fc_1, pred])

                program = cost.block.program
                with framework.program_guard(program, None):
                    optimize_ops = sgd.apply_gradients(params_grads)

                print("Finished apply gradients")
        """

        return self._optimizer.apply_gradients(params_grads=params_grads)

    def backward(self,
                 loss,
                 startup_program=None,
                 parameter_list=None,
                 no_grad_set=None,
                 callbacks=None,
                 checkpoints=None):
        """
        call append_backward with checkpoints.

        Args:
            loss (Variable): loss variable to run optimizations.
            startup_program (Program): startup_program for initializing parameters
                in `parameter_list`.
3855 3856
            parameter_list (list): list of Variables or Variable.names to update.
            no_grad_set (set|None): set of Variables or Variables.names should be ignored.
M
mapingshuo 已提交
3857 3858 3859 3860 3861 3862 3863 3864 3865 3866 3867 3868 3869 3870 3871 3872 3873 3874 3875 3876 3877 3878 3879 3880 3881 3882 3883 3884 3885 3886 3887 3888 3889 3890 3891 3892 3893 3894 3895 3896 3897 3898 3899 3900 3901
            callbacks (list|None): list of callables to run when appending backward
                operator for one parameter.
            checkpoints (list): list of Variables as checkpoints

        Examples:
            .. code-block:: python

                import paddle.fluid as fluid
    
                def mlp(input_x, input_y, hid_dim=128, label_dim=2):
                    fc_1 = fluid.layers.fc(input=input_x, size=hid_dim)
                    prediction = fluid.layers.fc(input=[fc_1], size=label_dim, act='softmax')
                    cost = fluid.layers.cross_entropy(input=prediction, label=input_y)
                    sum_cost = fluid.layers.reduce_mean(cost)
                    return sum_cost, fc_1, prediction
    
    
                input_x = fluid.layers.data(name="x", shape=[32], dtype='float32')
                input_y = fluid.layers.data(name="y", shape=[1], dtype='int64')
                cost, fc_1, pred = mlp(input_x, input_y)
                print("Finished FF")
    
                sgd = fluid.optimizer.Adam(learning_rate=0.01)
                sgd = fluid.optimizer.RecomputeOptimizer(sgd)
                params_grads = sgd.backward(
                    cost,
                    startup_program=None,
                    parameter_list=None,
                    no_grad_set=None,
                    checkpoints=[fc_1, pred])
                print("Finished backward")
        """

        if framework.in_dygraph_mode():
            raise NotImplementedError(
                "DyGraph current does not support recompute")

        self._dtype = loss.dtype
        program = loss.block.program
        with program_guard(program, startup_program):
            params_grads = append_backward(
                loss,
                parameter_list,
                no_grad_set,
                checkpoints=self._checkpoints)
3902 3903 3904
            # Note: since we can't use all_reduce_op now,
            #  dgc_op should be the last op of one grad.
            self._optimizer._append_dgc_ops(params_grads)
M
mapingshuo 已提交
3905 3906 3907 3908 3909 3910 3911 3912 3913 3914 3915 3916 3917 3918
        return params_grads

    def apply_optimize(self, loss, startup_program, params_grads):
        """
        call the apply_optimize function of self._optimizer

        Args:
            loss (Variable): loss variable to run optimizations.
            startup_program (Program): startup_program for initializing parameters
                in `parameter_list`.
            params_grads (list): list of (param, grad) pair to do optimization.

        Examples:
            .. code-block:: python
M
mapingshuo 已提交
3919

M
mapingshuo 已提交
3920 3921 3922 3923 3924 3925 3926
                import paddle.fluid as fluid
                
                def mlp(input_x, input_y, hid_dim=128, label_dim=2):
                    fc_1 = fluid.layers.fc(input=input_x, size=hid_dim)
                    prediction = fluid.layers.fc(input=[fc_1], size=label_dim, act='softmax')
                    cost = fluid.layers.cross_entropy(input=prediction, label=input_y)
                    sum_cost = fluid.layers.reduce_mean(cost)
M
mapingshuo 已提交
3927
                    return sum_cost, fc_1, prediction                
M
mapingshuo 已提交
3928 3929 3930 3931 3932 3933 3934 3935 3936 3937 3938 3939 3940 3941 3942 3943 3944 3945 3946
                
                input_x = fluid.layers.data(name="x", shape=[32], dtype='float32')
                input_y = fluid.layers.data(name="y", shape=[1], dtype='int64')
                cost, fc_1, pred = mlp(input_x, input_y)
                print("Finished FF")
                
                sgd = fluid.optimizer.Adam(learning_rate=0.01)
                sgd = fluid.optimizer.RecomputeOptimizer(sgd)
                params_grads = sgd.backward(
                    cost,
                    startup_program=None,
                    parameter_list=None,
                    no_grad_set=None,
                    checkpoints=[fc_1, pred])
                
                optimize_ops = sgd.apply_optimize(
                    cost, startup_program=None, params_grads=params_grads)
                
                print("Finished apply_optimize")
M
mapingshuo 已提交
3947

M
mapingshuo 已提交
3948 3949 3950 3951 3952 3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963 3964 3965 3966 3967 3968 3969 3970 3971 3972 3973 3974 3975 3976 3977 3978 3979 3980 3981 3982 3983
        """

        return self._optimizer.apply_optimize(
            loss, startup_program=startup_program, params_grads=params_grads)

    def minimize(self,
                 loss,
                 startup_program=None,
                 parameter_list=None,
                 no_grad_set=None,
                 grad_clip=None):

        assert (isinstance(loss, Variable)), "The loss should be an Variable."
        assert (self._checkpoints is not None
                ), "You should call _set_checkpoints first"
        if framework.in_dygraph_mode():
            raise NotImplementedError(
                "DyGraph current does not support recompute")

        params_grads = self.backward(
            loss,
            startup_program=startup_program,
            parameter_list=parameter_list,
            no_grad_set=no_grad_set,
            checkpoints=self._checkpoints)

        if grad_clip:
            # TODO(guru4elephant): should add grad_clip for static graph
            pass

        optimize_ops = self.apply_optimize(
            loss, startup_program=startup_program, params_grads=params_grads)

        return optimize_ops, params_grads


M
mapingshuo 已提交
3984 3985 3986 3987 3988 3989 3990 3991 3992 3993 3994 3995 3996 3997 3998 3999 4000 4001 4002 4003 4004 4005 4006 4007 4008 4009 4010 4011 4012 4013 4014 4015 4016 4017 4018 4019 4020 4021 4022 4023 4024 4025 4026 4027 4028 4029 4030 4031 4032 4033 4034 4035 4036 4037 4038
class LookaheadOptimizer(object):
    """
    This implements the Lookahead optimizer of the
    paper : https://arxiv.org/abs/1907.08610.

    Lookahead keeps two sets of params: the fast_params and
    the slow_params. inner_optimizer update fast_params every 
    training step. Lookahead updates the slow_params and fast_params 
    every k training steps as follows:

    .. math::
        
        slow\_param_t &= slow\_param_{t-1} + \\alpha * (fast\_param_{t-1} - slow\_param_{t-1})
	
	fast\_param_t &=  slow\_param_t

    Args:
        inner_optimizer (Optimizer): The optimizer that update fast params step by step. 
        alpha (float): The learning rate of Lookahead.
        k (int): The slow params is updated every k steps.

    Examples:
        .. code-block:: python

            import paddle
            import paddle.fluid as fluid
            import numpy as np

	    x = fluid.layers.data(name='x', shape=[2], dtype='float32')
	    label = fluid.layers.data(name="label", shape=[1], dtype="int64")
	    y = fluid.layers.fc(input=[x], size=2, act="softmax")
	    loss = fluid.layers.cross_entropy(input=y, label=label)
	    loss = fluid.layers.mean(x=loss)
	    sgd = fluid.optimizer.SGD(learning_rate=0.01)
	    optimizer = fluid.optimizer.LookaheadOptimizer(sgd,
                                            alpha=0.5,
                                            k=5)
	    optimizer.minimize(loss)
	    main_program = fluid.default_main_program()
	    place = fluid.CPUPlace()
	    exe = fluid.Executor(place)
	    exe.run(fluid.default_startup_program())

	    feeder = fluid.DataFeeder(feed_list=[x, label], place=place)

	    step = 0
            while(step < 10):
                step += 1
		exe.run(fluid.default_main_program(),
            	feed=feeder.feed(batch_data))

    """

    def __init__(self, inner_optimizer, alpha=0.5, k=5):

Z
zhongpu 已提交
4039 4040
        if framework.in_dygraph_mode():
            raise Exception("In dygraph, don't support LookaheadOptimizer.")
M
mapingshuo 已提交
4041 4042 4043 4044 4045 4046 4047 4048 4049 4050 4051 4052 4053 4054 4055 4056 4057 4058 4059 4060 4061 4062 4063 4064 4065 4066 4067 4068 4069 4070 4071 4072 4073 4074 4075 4076 4077 4078 4079 4080 4081 4082 4083 4084 4085 4086 4087 4088 4089 4090 4091 4092 4093 4094 4095 4096 4097 4098 4099 4100 4101 4102 4103 4104 4105 4106 4107 4108 4109 4110 4111 4112 4113 4114 4115 4116 4117 4118 4119 4120 4121 4122 4123 4124 4125 4126 4127 4128 4129 4130 4131 4132 4133 4134 4135 4136
        assert (inner_optimizer is not None), "inner optimizer can not be None"
        assert (
            0.0 <= alpha <= 1.0
        ), "alpha should be larger or equal to 0.0, and less or equal than 1.0"
        assert (isinstance(k, int) and k > 0), "k should be a positive integer"

        self.inner_optimizer = inner_optimizer
        self.alpha = alpha
        self.k = k
        self.type = "lookahead"

    def minimize(self, loss, startup_program=None):

        # Apply inner optimizer to the main_program
        mini_out = self.inner_optimizer.minimize(
            loss, startup_program=startup_program)

        # Get startup_program and main_program
        if startup_program is None:
            startup_program = default_startup_program()
        main_block = loss.block

        # add some vars to the main_program
        params = [param.name for param in main_block.all_parameters()]
        param_to_slow = {}
        for param in params:
            fast_var = main_block.var(param)
            assert (fast_var is not None)
            slow_var = main_block.create_var(
                name=param + "@SLOW",
                shape=fast_var.shape,
                dtype=fast_var.dtype,
                persistable=True)
            param_to_slow[param] = slow_var

        # add some vars to the startup_program
        startup_block = startup_program.global_block()
        for param in params:
            fast_var = startup_block.var(param)
            assert (fast_var is not None)
            slow_var = startup_block.create_var(
                name=param + "@SLOW",
                shape=fast_var.shape,
                dtype=fast_var.dtype,
                persistable=True)

            startup_block.append_op(
                type="assign",
                inputs={"X": fast_var},
                outputs={"Out": slow_var})

        # Add Var k to main prog and startup prog
        k = layers.create_global_var(
            name="lookahead_k",
            shape=[1],
            value=int(self.k),
            dtype='int32',
            persistable=True)

        # Add Var alpha to main prog and startup prog
        alpha = layers.create_global_var(
            name="lookahead_alpha",
            shape=[1],
            value=float(self.alpha),
            dtype='float32',
            persistable=True)

        # Add Var step
        step = layers.create_global_var(
            name="lookahead_step",
            shape=[1],
            value=int(0),
            dtype='int32',
            persistable=True)
        layers.increment(x=step, value=1.0, in_place=True)

        # lookahead
        zero_var = layers.fill_constant(shape=[1], dtype='float32', value=0.0)

        one_var = layers.fill_constant(shape=[1], dtype='float32', value=1.0)

        mod = layers.elementwise_mod(step, k)
        with layers.control_flow.Switch() as switch:
            with switch.case(mod == zero_var):
                for param_name in params:
                    fast_var = main_block.var(param_name)
                    slow_var = param_to_slow[param_name]
                    tmp_var = layers.elementwise_add(
                        layers.elementwise_mul(fast_var, alpha),
                        layers.elementwise_mul(
                            slow_var, layers.elementwise_sub(one_var, alpha)))
                    layers.assign(input=tmp_var, output=slow_var)
                    layers.assign(input=tmp_var, output=fast_var)
            with switch.default():
                pass
        return mini_out